Index: head/sys/fs/nfs/nfs_commonkrpc.c =================================================================== --- head/sys/fs/nfs/nfs_commonkrpc.c (revision 321627) +++ head/sys/fs/nfs/nfs_commonkrpc.c (revision 321628) @@ -1,1325 +1,1325 @@ /*- * Copyright (c) 1989, 1991, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); /* * Socket operations for use by nfs */ #include "opt_kgssapi.h" #include "opt_nfs.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_nfsclient_nfs23_start_probe_func_t dtrace_nfscl_nfs234_start_probe; dtrace_nfsclient_nfs23_done_probe_func_t dtrace_nfscl_nfs234_done_probe; /* * Registered probes by RPC type. */ uint32_t nfscl_nfs2_start_probes[NFSV41_NPROCS + 1]; uint32_t nfscl_nfs2_done_probes[NFSV41_NPROCS + 1]; uint32_t nfscl_nfs3_start_probes[NFSV41_NPROCS + 1]; uint32_t nfscl_nfs3_done_probes[NFSV41_NPROCS + 1]; uint32_t nfscl_nfs4_start_probes[NFSV41_NPROCS + 1]; uint32_t nfscl_nfs4_done_probes[NFSV41_NPROCS + 1]; #endif NFSSTATESPINLOCK; NFSREQSPINLOCK; NFSDLOCKMUTEX; NFSCLSTATEMUTEX; extern struct nfsstatsv1 nfsstatsv1; extern struct nfsreqhead nfsd_reqq; extern int nfscl_ticks; extern void (*ncl_call_invalcaches)(struct vnode *); extern int nfs_numnfscbd; extern int nfscl_debuglevel; SVCPOOL *nfscbd_pool; static int nfsrv_gsscallbackson = 0; static int nfs_bufpackets = 4; static int nfs_reconnects; static int nfs3_jukebox_delay = 10; static int nfs_skip_wcc_data_onerr = 1; SYSCTL_DECL(_vfs_nfs); SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "Buffer reservation size 2 < x < 64"); SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0, "Number of times the nfs client has had to reconnect"); SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0, "Number of seconds to delay a retry after receiving EJUKEBOX"); SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0, "Disable weak cache consistency checking when server returns an error"); static void nfs_down(struct nfsmount *, struct thread *, const char *, int, int); static void nfs_up(struct nfsmount *, struct thread *, const char *, int, int); static int nfs_msg(struct thread *, const char *, const char *, int); struct nfs_cached_auth { int ca_refs; /* refcount, including 1 from the cache */ uid_t ca_uid; /* uid that corresponds to this auth */ AUTH *ca_auth; /* RPC auth handle */ }; static int nfsv2_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, }; /* * Initialize sockets and congestion for a new NFS connection. * We do not free the sockaddr if error. */ int newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp, struct ucred *cred, NFSPROC_T *p, int callback_retry_mult) { int rcvreserve, sndreserve; int pktscale, pktscalesav; struct sockaddr *saddr; struct ucred *origcred; CLIENT *client; struct netconfig *nconf; struct socket *so; int one = 1, retries, error = 0; struct thread *td = curthread; SVCXPRT *xprt; struct timeval timo; /* * We need to establish the socket using the credentials of * the mountpoint. Some parts of this process (such as * sobind() and soconnect()) will use the curent thread's * credential instead of the socket credential. To work * around this, temporarily change the current thread's * credential to that of the mountpoint. * * XXX: It would be better to explicitly pass the correct * credential to sobind() and soconnect(). */ origcred = td->td_ucred; /* * Use the credential in nr_cred, if not NULL. */ if (nrp->nr_cred != NULL) td->td_ucred = nrp->nr_cred; else td->td_ucred = cred; saddr = nrp->nr_nam; if (saddr->sa_family == AF_INET) if (nrp->nr_sotype == SOCK_DGRAM) nconf = getnetconfigent("udp"); else nconf = getnetconfigent("tcp"); else if (saddr->sa_family == AF_LOCAL) nconf = getnetconfigent("local"); else if (nrp->nr_sotype == SOCK_DGRAM) nconf = getnetconfigent("udp6"); else nconf = getnetconfigent("tcp6"); pktscale = nfs_bufpackets; if (pktscale < 2) pktscale = 2; if (pktscale > 64) pktscale = 64; pktscalesav = pktscale; /* * soreserve() can fail if sb_max is too small, so shrink pktscale * and try again if there is an error. * Print a log message suggesting increasing sb_max. * Creating a socket and doing this is necessary since, if the * reservation sizes are too large and will make soreserve() fail, * the connection will work until a large send is attempted and * then it will loop in the krpc code. */ so = NULL; saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *); error = socreate(saddr->sa_family, &so, nrp->nr_sotype, nrp->nr_soproto, td->td_ucred, td); if (error) { td->td_ucred = origcred; goto out; } do { if (error != 0 && pktscale > 2) { if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM && pktscale == pktscalesav) printf("Consider increasing kern.ipc.maxsockbuf\n"); pktscale--; } if (nrp->nr_sotype == SOCK_DGRAM) { if (nmp != NULL) { sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) * pktscale; rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) * pktscale; } else { sndreserve = rcvreserve = 1024 * pktscale; } } else { if (nrp->nr_sotype != SOCK_STREAM) panic("nfscon sotype"); if (nmp != NULL) { sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR + sizeof (u_int32_t)) * pktscale; rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR + sizeof (u_int32_t)) * pktscale; } else { sndreserve = rcvreserve = 1024 * pktscale; } } error = soreserve(so, sndreserve, rcvreserve); if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM && pktscale <= 2) printf("Must increase kern.ipc.maxsockbuf or reduce" " rsize, wsize\n"); } while (error != 0 && pktscale > 2); soclose(so); if (error) { td->td_ucred = origcred; goto out; } client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog, nrp->nr_vers, sndreserve, rcvreserve); CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq"); if (nmp != NULL) { if ((nmp->nm_flag & NFSMNT_INT)) CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one); if ((nmp->nm_flag & NFSMNT_RESVPORT)) CLNT_CONTROL(client, CLSET_PRIVPORT, &one); if (NFSHASSOFT(nmp)) { if (nmp->nm_sotype == SOCK_DGRAM) /* * For UDP, the large timeout for a reconnect * will be set to "nm_retry * nm_timeo / 2", so * we only want to do 2 reconnect timeout * retries. */ retries = 2; else retries = nmp->nm_retry; } else retries = INT_MAX; /* cred == NULL for DS connects. */ if (NFSHASNFSV4N(nmp) && cred != NULL) { /* * Make sure the nfscbd_pool doesn't get destroyed * while doing this. */ NFSD_LOCK(); if (nfs_numnfscbd > 0) { nfs_numnfscbd++; NFSD_UNLOCK(); xprt = svc_vc_create_backchannel(nfscbd_pool); CLNT_CONTROL(client, CLSET_BACKCHANNEL, xprt); NFSD_LOCK(); nfs_numnfscbd--; if (nfs_numnfscbd == 0) wakeup(&nfs_numnfscbd); } NFSD_UNLOCK(); } } else { /* * Three cases: * - Null RPC callback to client * - Non-Null RPC callback to client, wait a little longer * - upcalls to nfsuserd and gssd (clp == NULL) */ if (callback_retry_mult == 0) { retries = NFSV4_UPCALLRETRY; CLNT_CONTROL(client, CLSET_PRIVPORT, &one); } else { retries = NFSV4_CALLBACKRETRY * callback_retry_mult; } } CLNT_CONTROL(client, CLSET_RETRIES, &retries); if (nmp != NULL) { /* * For UDP, there are 2 timeouts: * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer * that does a retransmit of an RPC request using the same * socket and xid. This is what you normally want to do, * since NFS servers depend on "same xid" for their * Duplicate Request Cache. * - timeout specified in CLNT_CALL_MBUF(), which specifies when * retransmits on the same socket should fail and a fresh * socket created. Each of these timeouts counts as one * CLSET_RETRIES as set above. * Set the initial retransmit timeout for UDP. This timeout * doesn't exist for TCP and the following call just fails, * which is ok. */ timo.tv_sec = nmp->nm_timeo / NFS_HZ; timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ; CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo); } mtx_lock(&nrp->nr_mtx); if (nrp->nr_client != NULL) { mtx_unlock(&nrp->nr_mtx); /* * Someone else already connected. */ CLNT_RELEASE(client); } else { nrp->nr_client = client; /* * Protocols that do not require connections may be optionally * left unconnected for servers that reply from a port other * than NFS_PORT. */ if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) { mtx_unlock(&nrp->nr_mtx); CLNT_CONTROL(client, CLSET_CONNECT, &one); } else mtx_unlock(&nrp->nr_mtx); } /* Restore current thread's credentials. */ td->td_ucred = origcred; out: NFSEXITCODE(error); return (error); } /* * NFS disconnect. Clean up and unlink. */ void newnfs_disconnect(struct nfssockreq *nrp) { CLIENT *client; mtx_lock(&nrp->nr_mtx); if (nrp->nr_client != NULL) { client = nrp->nr_client; nrp->nr_client = NULL; mtx_unlock(&nrp->nr_mtx); rpc_gss_secpurge_call(client); CLNT_CLOSE(client); CLNT_RELEASE(client); } else { mtx_unlock(&nrp->nr_mtx); } } static AUTH * nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal, char *srv_principal, gss_OID mech_oid, struct ucred *cred) { rpc_gss_service_t svc; AUTH *auth; switch (secflavour) { case RPCSEC_GSS_KRB5: case RPCSEC_GSS_KRB5I: case RPCSEC_GSS_KRB5P: if (!mech_oid) { if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid)) return (NULL); } if (secflavour == RPCSEC_GSS_KRB5) svc = rpc_gss_svc_none; else if (secflavour == RPCSEC_GSS_KRB5I) svc = rpc_gss_svc_integrity; else svc = rpc_gss_svc_privacy; if (clnt_principal == NULL) auth = rpc_gss_secfind_call(nrp->nr_client, cred, srv_principal, mech_oid, svc); else { auth = rpc_gss_seccreate_call(nrp->nr_client, cred, clnt_principal, srv_principal, "kerberosv5", svc, NULL, NULL, NULL); return (auth); } if (auth != NULL) return (auth); /* fallthrough */ case AUTH_SYS: default: return (authunix_create(cred)); } } /* * Callback from the RPC code to generate up/down notifications. */ struct nfs_feedback_arg { struct nfsmount *nf_mount; int nf_lastmsg; /* last tprintf */ int nf_tprintfmsg; struct thread *nf_td; }; static void nfs_feedback(int type, int proc, void *arg) { struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg; struct nfsmount *nmp = nf->nf_mount; time_t now; switch (type) { case FEEDBACK_REXMIT2: case FEEDBACK_RECONNECT: now = NFSD_MONOSEC; if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) { nfs_down(nmp, nf->nf_td, "not responding", 0, NFSSTA_TIMEO); nf->nf_tprintfmsg = TRUE; nf->nf_lastmsg = now; } break; case FEEDBACK_OK: nfs_up(nf->nf_mount, nf->nf_td, "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg); break; } } /* * newnfs_request - goes something like this * - does the rpc by calling the krpc layer * - break down rpc header and return with nfs reply * nb: always frees up nd_mreq mbuf list */ int newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp, struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp, struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers, u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep) { uint32_t retseq, retval, slotseq, *tl; time_t waituntil; int i = 0, j = 0, opcnt, set_sigset = 0, slot; int trycnt, error = 0, usegssname = 0, secflavour = AUTH_SYS; int freeslot, maxslot, reterr, slotpos, timeo; u_int16_t procnum; u_int trylater_delay = 1; struct nfs_feedback_arg nf; struct timeval timo; AUTH *auth; struct rpc_callextra ext; enum clnt_stat stat; struct nfsreq *rep = NULL; char *srv_principal = NULL, *clnt_principal = NULL; sigset_t oldset; struct ucred *authcred; struct nfsclsession *sep; uint8_t sessionid[NFSX_V4SESSIONID]; sep = dssep; if (xidp != NULL) *xidp = 0; /* Reject requests while attempting a forced unmount. */ - if (nmp != NULL && (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) { + if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) { m_freem(nd->nd_mreq); return (ESTALE); } /* * Set authcred, which is used to acquire RPC credentials to * the cred argument, by default. The crhold() should not be * necessary, but will ensure that some future code change * doesn't result in the credential being free'd prematurely. */ authcred = crhold(cred); /* For client side interruptible mounts, mask off the signals. */ if (nmp != NULL && td != NULL && NFSHASINT(nmp)) { newnfs_set_sigmask(td, &oldset); set_sigset = 1; } /* * XXX if not already connected call nfs_connect now. Longer * term, change nfs_mount to call nfs_connect unconditionally * and let clnt_reconnect_create handle reconnects. */ if (nrp->nr_client == NULL) newnfs_connect(nmp, nrp, cred, td, 0); /* * For a client side mount, nmp is != NULL and clp == NULL. For * server calls (callbacks or upcalls), nmp == NULL. */ if (clp != NULL) { NFSLOCKSTATE(); if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) { secflavour = RPCSEC_GSS_KRB5; if (nd->nd_procnum != NFSPROC_NULL) { if (clp->lc_flags & LCL_GSSINTEGRITY) secflavour = RPCSEC_GSS_KRB5I; else if (clp->lc_flags & LCL_GSSPRIVACY) secflavour = RPCSEC_GSS_KRB5P; } } NFSUNLOCKSTATE(); } else if (nmp != NULL && NFSHASKERB(nmp) && nd->nd_procnum != NFSPROC_NULL) { if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0) nd->nd_flag |= ND_USEGSSNAME; if ((nd->nd_flag & ND_USEGSSNAME) != 0) { /* * If there is a client side host based credential, * use that, otherwise use the system uid, if set. * The system uid is in the nmp->nm_sockreq.nr_cred * credentials. */ if (nmp->nm_krbnamelen > 0) { usegssname = 1; clnt_principal = nmp->nm_krbname; } else if (nmp->nm_uid != (uid_t)-1) { KASSERT(nmp->nm_sockreq.nr_cred != NULL, ("newnfs_request: NULL nr_cred")); crfree(authcred); authcred = crhold(nmp->nm_sockreq.nr_cred); } } else if (nmp->nm_krbnamelen == 0 && nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) { /* * If there is no host based principal name and * the system uid is set and this is root, use the * system uid, since root won't have user * credentials in a credentials cache file. * The system uid is in the nmp->nm_sockreq.nr_cred * credentials. */ KASSERT(nmp->nm_sockreq.nr_cred != NULL, ("newnfs_request: NULL nr_cred")); crfree(authcred); authcred = crhold(nmp->nm_sockreq.nr_cred); } if (NFSHASINTEGRITY(nmp)) secflavour = RPCSEC_GSS_KRB5I; else if (NFSHASPRIVACY(nmp)) secflavour = RPCSEC_GSS_KRB5P; else secflavour = RPCSEC_GSS_KRB5; srv_principal = NFSMNT_SRVKRBNAME(nmp); } else if (nmp != NULL && !NFSHASKERB(nmp) && nd->nd_procnum != NFSPROC_NULL && (nd->nd_flag & ND_USEGSSNAME) != 0) { /* * Use the uid that did the mount when the RPC is doing * NFSv4 system operations, as indicated by the * ND_USEGSSNAME flag, for the AUTH_SYS case. * The credentials in nm_sockreq.nr_cred were used for the * mount. */ KASSERT(nmp->nm_sockreq.nr_cred != NULL, ("newnfs_request: NULL nr_cred")); crfree(authcred); authcred = crhold(nmp->nm_sockreq.nr_cred); } if (nmp != NULL) { bzero(&nf, sizeof(struct nfs_feedback_arg)); nf.nf_mount = nmp; nf.nf_td = td; nf.nf_lastmsg = NFSD_MONOSEC - ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay)); } if (nd->nd_procnum == NFSPROC_NULL) auth = authnone_create(); else if (usegssname) { /* * For this case, the authenticator is held in the * nfssockreq structure, so don't release the reference count * held on it. --> Don't AUTH_DESTROY() it in this function. */ if (nrp->nr_auth == NULL) nrp->nr_auth = nfs_getauth(nrp, secflavour, clnt_principal, srv_principal, NULL, authcred); else rpc_gss_refresh_auth_call(nrp->nr_auth); auth = nrp->nr_auth; } else auth = nfs_getauth(nrp, secflavour, NULL, srv_principal, NULL, authcred); crfree(authcred); if (auth == NULL) { m_freem(nd->nd_mreq); if (set_sigset) newnfs_restore_sigmask(td, &oldset); return (EACCES); } bzero(&ext, sizeof(ext)); ext.rc_auth = auth; if (nmp != NULL) { ext.rc_feedback = nfs_feedback; ext.rc_feedback_arg = &nf; } procnum = nd->nd_procnum; if ((nd->nd_flag & ND_NFSV4) && nd->nd_procnum != NFSPROC_NULL && nd->nd_procnum != NFSV4PROC_CBCOMPOUND) procnum = NFSV4PROC_COMPOUND; if (nmp != NULL) { NFSINCRGLOBAL(nfsstatsv1.rpcrequests); /* Map the procnum to the old NFSv2 one, as required. */ if ((nd->nd_flag & ND_NFSV2) != 0) { if (nd->nd_procnum < NFS_V3NPROCS) procnum = nfsv2_procid[nd->nd_procnum]; else procnum = NFSV2PROC_NOOP; } /* * Now only used for the R_DONTRECOVER case, but until that is * supported within the krpc code, I need to keep a queue of * outstanding RPCs for nfsv4 client requests. */ if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND) MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSDREQ, M_WAITOK); #ifdef KDTRACE_HOOKS if (dtrace_nfscl_nfs234_start_probe != NULL) { uint32_t probe_id; int probe_procnum; if (nd->nd_flag & ND_NFSV4) { probe_id = nfscl_nfs4_start_probes[nd->nd_procnum]; probe_procnum = nd->nd_procnum; } else if (nd->nd_flag & ND_NFSV3) { probe_id = nfscl_nfs3_start_probes[procnum]; probe_procnum = procnum; } else { probe_id = nfscl_nfs2_start_probes[nd->nd_procnum]; probe_procnum = procnum; } if (probe_id != 0) (dtrace_nfscl_nfs234_start_probe) (probe_id, vp, nd->nd_mreq, cred, probe_procnum); } #endif } trycnt = 0; freeslot = -1; /* Set to slot that needs to be free'd */ tryagain: slot = -1; /* Slot that needs a sequence# increment. */ /* * This timeout specifies when a new socket should be created, * along with new xid values. For UDP, this should be done * infrequently, since retransmits of RPC requests should normally * use the same xid. */ if (nmp == NULL) { timo.tv_usec = 0; if (clp == NULL) timo.tv_sec = NFSV4_UPCALLTIMEO; else timo.tv_sec = NFSV4_CALLBACKTIMEO; } else { if (nrp->nr_sotype != SOCK_DGRAM) { timo.tv_usec = 0; if ((nmp->nm_flag & NFSMNT_NFSV4)) timo.tv_sec = INT_MAX; else timo.tv_sec = NFS_TCPTIMEO; } else { if (NFSHASSOFT(nmp)) { /* * CLSET_RETRIES is set to 2, so this should be * half of the total timeout required. */ timeo = nmp->nm_retry * nmp->nm_timeo / 2; if (timeo < 1) timeo = 1; timo.tv_sec = timeo / NFS_HZ; timo.tv_usec = (timeo % NFS_HZ) * 1000000 / NFS_HZ; } else { /* For UDP hard mounts, use a large value. */ timo.tv_sec = NFS_MAXTIMEO / NFS_HZ; timo.tv_usec = 0; } } if (rep != NULL) { rep->r_flags = 0; rep->r_nmp = nmp; /* * Chain request into list of outstanding requests. */ NFSLOCKREQ(); TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain); NFSUNLOCKREQ(); } } nd->nd_mrep = NULL; if (clp != NULL && sep != NULL) stat = clnt_bck_call(nrp->nr_client, &ext, procnum, nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt); else stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum, nd->nd_mreq, &nd->nd_mrep, timo); if (rep != NULL) { /* * RPC done, unlink the request. */ NFSLOCKREQ(); TAILQ_REMOVE(&nfsd_reqq, rep, r_chain); NFSUNLOCKREQ(); } /* * If there was a successful reply and a tprintf msg. * tprintf a response. */ if (stat == RPC_SUCCESS) { error = 0; } else if (stat == RPC_TIMEDOUT) { NFSINCRGLOBAL(nfsstatsv1.rpctimeouts); error = ETIMEDOUT; } else if (stat == RPC_VERSMISMATCH) { NFSINCRGLOBAL(nfsstatsv1.rpcinvalid); error = EOPNOTSUPP; } else if (stat == RPC_PROGVERSMISMATCH) { NFSINCRGLOBAL(nfsstatsv1.rpcinvalid); error = EPROTONOSUPPORT; } else if (stat == RPC_INTR) { error = EINTR; } else { NFSINCRGLOBAL(nfsstatsv1.rpcinvalid); error = EACCES; } if (error) { m_freem(nd->nd_mreq); if (usegssname == 0) AUTH_DESTROY(auth); if (rep != NULL) FREE((caddr_t)rep, M_NFSDREQ); if (set_sigset) newnfs_restore_sigmask(td, &oldset); return (error); } KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n")); /* * Search for any mbufs that are not a multiple of 4 bytes long * or with m_data not longword aligned. * These could cause pointer alignment problems, so copy them to * well aligned mbufs. */ newnfs_realign(&nd->nd_mrep, M_WAITOK); nd->nd_md = nd->nd_mrep; nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t); nd->nd_repstat = 0; if (nd->nd_procnum != NFSPROC_NULL && nd->nd_procnum != NFSV4PROC_CBNULL) { /* If sep == NULL, set it to the default in nmp. */ if (sep == NULL && nmp != NULL) sep = nfsmnt_mdssession(nmp); /* * and now the actual NFS xdr. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl); if (nd->nd_repstat >= 10000) NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum, (int)nd->nd_repstat); /* * Get rid of the tag, return count and SEQUENCE result for * NFSv4. */ if ((nd->nd_flag & ND_NFSV4) != 0) { 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; NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); opcnt = fxdr_unsigned(int, *tl++); i = fxdr_unsigned(int, *tl++); j = fxdr_unsigned(int, *tl); if (j >= 10000) NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j); /* * If the first op is Sequence, free up the slot. */ if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) || (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0)) NFSCL_DEBUG(1, "failed seq=%d\n", j); if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) || (clp != NULL && i == NFSV4OP_CBSEQUENCE && j == 0) ) { if (i == NFSV4OP_SEQUENCE) NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED); else NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED); mtx_lock(&sep->nfsess_mtx); if (bcmp(tl, sep->nfsess_sessionid, NFSX_V4SESSIONID) == 0) { tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; retseq = fxdr_unsigned(uint32_t, *tl++); slot = fxdr_unsigned(int, *tl++); freeslot = slot; if (retseq != sep->nfsess_slotseq[slot]) printf("retseq diff 0x%x\n", retseq); retval = fxdr_unsigned(uint32_t, *++tl); if ((retval + 1) < sep->nfsess_foreslots ) sep->nfsess_foreslots = (retval + 1); else if ((retval + 1) > sep->nfsess_foreslots) sep->nfsess_foreslots = (retval < 64) ? (retval + 1) : 64; } mtx_unlock(&sep->nfsess_mtx); /* Grab the op and status for the next one. */ if (opcnt > 1) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl++); j = fxdr_unsigned(int, *tl); } } } if (nd->nd_repstat != 0) { if (nd->nd_repstat == NFSERR_BADSESSION && nmp != NULL && dssep == NULL) { /* * If this is a client side MDS RPC, mark * the MDS session defunct and initiate * recovery, as required. * The nfsess_defunct field is protected by * the NFSLOCKMNT()/nm_mtx lock and not the * nfsess_mtx lock to simplify its handling, * for the MDS session. This lock is also * sufficient for nfsess_sessionid, since it * never changes in the structure. */ NFSCL_DEBUG(1, "Got badsession\n"); NFSLOCKCLSTATE(); NFSLOCKMNT(nmp); sep = NFSMNT_MDSSESSION(nmp); if (bcmp(sep->nfsess_sessionid, nd->nd_sequence, NFSX_V4SESSIONID) == 0) { /* Initiate recovery. */ sep->nfsess_defunct = 1; NFSCL_DEBUG(1, "Marked defunct\n"); if (nmp->nm_clp != NULL) { nmp->nm_clp->nfsc_flags |= NFSCLFLAGS_RECOVER; wakeup(nmp->nm_clp); } } NFSUNLOCKCLSTATE(); /* * Sleep for up to 1sec waiting for a new * session. */ mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO, "nfsbadsess", hz); /* * Get the session again, in case a new one * has been created during the sleep. */ sep = NFSMNT_MDSSESSION(nmp); NFSUNLOCKMNT(nmp); if ((nd->nd_flag & ND_LOOPBADSESS) != 0) { reterr = nfsv4_sequencelookup(nmp, sep, &slotpos, &maxslot, &slotseq, sessionid); if (reterr == 0) { /* Fill in new session info. */ NFSCL_DEBUG(1, "Filling in new sequence\n"); tl = nd->nd_sequence; bcopy(sessionid, tl, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; *tl++ = txdr_unsigned(slotseq); *tl++ = txdr_unsigned(slotpos); *tl = txdr_unsigned(maxslot); } if (reterr == NFSERR_BADSESSION || reterr == 0) { NFSCL_DEBUG(1, "Badsession looping\n"); m_freem(nd->nd_mrep); nd->nd_mrep = NULL; goto tryagain; } nd->nd_repstat = reterr; NFSCL_DEBUG(1, "Got err=%d\n", reterr); } } if (((nd->nd_repstat == NFSERR_DELAY || nd->nd_repstat == NFSERR_GRACE) && (nd->nd_flag & ND_NFSV4) && nd->nd_procnum != NFSPROC_DELEGRETURN && nd->nd_procnum != NFSPROC_SETATTR && nd->nd_procnum != NFSPROC_READ && nd->nd_procnum != NFSPROC_READDS && nd->nd_procnum != NFSPROC_WRITE && nd->nd_procnum != NFSPROC_WRITEDS && nd->nd_procnum != NFSPROC_OPEN && nd->nd_procnum != NFSPROC_CREATE && nd->nd_procnum != NFSPROC_OPENCONFIRM && nd->nd_procnum != NFSPROC_OPENDOWNGRADE && nd->nd_procnum != NFSPROC_CLOSE && nd->nd_procnum != NFSPROC_LOCK && nd->nd_procnum != NFSPROC_LOCKU) || (nd->nd_repstat == NFSERR_DELAY && (nd->nd_flag & ND_NFSV4) == 0) || nd->nd_repstat == NFSERR_RESOURCE) { if (trylater_delay > NFS_TRYLATERDEL) trylater_delay = NFS_TRYLATERDEL; waituntil = NFSD_MONOSEC + trylater_delay; while (NFSD_MONOSEC < waituntil) (void) nfs_catnap(PZERO, 0, "nfstry"); trylater_delay *= 2; if (slot != -1) { mtx_lock(&sep->nfsess_mtx); sep->nfsess_slotseq[slot]++; *nd->nd_slotseq = txdr_unsigned( sep->nfsess_slotseq[slot]); mtx_unlock(&sep->nfsess_mtx); } m_freem(nd->nd_mrep); nd->nd_mrep = NULL; goto tryagain; } /* * If the File Handle was stale, invalidate the * lookup cache, just in case. * (vp != NULL implies a client side call) */ if (nd->nd_repstat == ESTALE && vp != NULL) { cache_purge(vp); if (ncl_call_invalcaches != NULL) (*ncl_call_invalcaches)(vp); } } if ((nd->nd_flag & ND_NFSV4) != 0) { /* Free the slot, as required. */ if (freeslot != -1) nfsv4_freeslot(sep, freeslot); /* * If this op is Putfh, throw its results away. */ if (j >= 10000) NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j); if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) { NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl++); j = fxdr_unsigned(int, *tl); if (j >= 10000) NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i, j); /* * All Compounds that do an Op that must * be in sequence consist of NFSV4OP_PUTFH * followed by one of these. As such, we * can determine if the seqid# should be * incremented, here. */ if ((i == NFSV4OP_OPEN || i == NFSV4OP_OPENCONFIRM || i == NFSV4OP_OPENDOWNGRADE || i == NFSV4OP_CLOSE || i == NFSV4OP_LOCK || i == NFSV4OP_LOCKU) && (j == 0 || (j != NFSERR_STALECLIENTID && j != NFSERR_STALESTATEID && j != NFSERR_BADSTATEID && j != NFSERR_BADSEQID && j != NFSERR_BADXDR && j != NFSERR_RESOURCE && j != NFSERR_NOFILEHANDLE))) nd->nd_flag |= ND_INCRSEQID; } /* * If this op's status is non-zero, mark * that there is no more data to process. * The exception is Setattr, which always has xdr * when it has failed. */ if (j != 0 && i != NFSV4OP_SETATTR) nd->nd_flag |= ND_NOMOREDATA; /* * If R_DONTRECOVER is set, replace the stale error * reply, so that recovery isn't initiated. */ if ((nd->nd_repstat == NFSERR_STALECLIENTID || nd->nd_repstat == NFSERR_BADSESSION || nd->nd_repstat == NFSERR_STALESTATEID) && rep != NULL && (rep->r_flags & R_DONTRECOVER)) nd->nd_repstat = NFSERR_STALEDONTRECOVER; } } #ifdef KDTRACE_HOOKS if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) { uint32_t probe_id; int probe_procnum; if (nd->nd_flag & ND_NFSV4) { probe_id = nfscl_nfs4_done_probes[nd->nd_procnum]; probe_procnum = nd->nd_procnum; } else if (nd->nd_flag & ND_NFSV3) { probe_id = nfscl_nfs3_done_probes[procnum]; probe_procnum = procnum; } else { probe_id = nfscl_nfs2_done_probes[nd->nd_procnum]; probe_procnum = procnum; } if (probe_id != 0) (dtrace_nfscl_nfs234_done_probe)(probe_id, vp, nd->nd_mreq, cred, probe_procnum, 0); } #endif m_freem(nd->nd_mreq); if (usegssname == 0) AUTH_DESTROY(auth); if (rep != NULL) FREE((caddr_t)rep, M_NFSDREQ); if (set_sigset) newnfs_restore_sigmask(td, &oldset); return (0); nfsmout: mbuf_freem(nd->nd_mrep); mbuf_freem(nd->nd_mreq); if (usegssname == 0) AUTH_DESTROY(auth); if (rep != NULL) FREE((caddr_t)rep, M_NFSDREQ); if (set_sigset) newnfs_restore_sigmask(td, &oldset); return (error); } /* * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and * wait for all requests to complete. This is used by forced unmounts * to terminate any outstanding RPCs. */ int newnfs_nmcancelreqs(struct nfsmount *nmp) { if (nmp->nm_sockreq.nr_client != NULL) CLNT_CLOSE(nmp->nm_sockreq.nr_client); return (0); } /* * Any signal that can interrupt an NFS operation in an intr mount * should be added to this set. SIGSTOP and SIGKILL cannot be masked. */ int newnfs_sig_set[] = { SIGINT, SIGTERM, SIGHUP, SIGKILL, SIGQUIT }; /* * Check to see if one of the signals in our subset is pending on * the process (in an intr mount). */ static int nfs_sig_pending(sigset_t set) { int i; for (i = 0 ; i < nitems(newnfs_sig_set); i++) if (SIGISMEMBER(set, newnfs_sig_set[i])) return (1); return (0); } /* * The set/restore sigmask functions are used to (temporarily) overwrite * the thread td_sigmask during an RPC call (for example). These are also * used in other places in the NFS client that might tsleep(). */ void newnfs_set_sigmask(struct thread *td, sigset_t *oldset) { sigset_t newset; int i; struct proc *p; SIGFILLSET(newset); if (td == NULL) td = curthread; /* XXX */ p = td->td_proc; /* Remove the NFS set of signals from newset */ PROC_LOCK(p); mtx_lock(&p->p_sigacts->ps_mtx); for (i = 0 ; i < nitems(newnfs_sig_set); i++) { /* * But make sure we leave the ones already masked * by the process, ie. remove the signal from the * temporary signalmask only if it wasn't already * in p_sigmask. */ if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) && !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i])) SIGDELSET(newset, newnfs_sig_set[i]); } mtx_unlock(&p->p_sigacts->ps_mtx); kern_sigprocmask(td, SIG_SETMASK, &newset, oldset, SIGPROCMASK_PROC_LOCKED); PROC_UNLOCK(p); } void newnfs_restore_sigmask(struct thread *td, sigset_t *set) { if (td == NULL) td = curthread; /* XXX */ kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0); } /* * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the * old one after msleep() returns. */ int newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo) { sigset_t oldset; int error; struct proc *p; if ((priority & PCATCH) == 0) return msleep(ident, mtx, priority, wmesg, timo); if (td == NULL) td = curthread; /* XXX */ newnfs_set_sigmask(td, &oldset); error = msleep(ident, mtx, priority, wmesg, timo); newnfs_restore_sigmask(td, &oldset); p = td->td_proc; return (error); } /* * Test for a termination condition pending on the process. * This is used for NFSMNT_INT mounts. */ int newnfs_sigintr(struct nfsmount *nmp, struct thread *td) { struct proc *p; sigset_t tmpset; /* Terminate all requests while attempting a forced unmount. */ - if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) + if (NFSCL_FORCEDISM(nmp->nm_mountp)) return (EIO); if (!(nmp->nm_flag & NFSMNT_INT)) return (0); if (td == NULL) return (0); p = td->td_proc; PROC_LOCK(p); tmpset = p->p_siglist; SIGSETOR(tmpset, td->td_siglist); SIGSETNAND(tmpset, td->td_sigmask); mtx_lock(&p->p_sigacts->ps_mtx); SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore); mtx_unlock(&p->p_sigacts->ps_mtx); if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist)) && nfs_sig_pending(tmpset)) { PROC_UNLOCK(p); return (EINTR); } PROC_UNLOCK(p); return (0); } static int nfs_msg(struct thread *td, const char *server, const char *msg, int error) { struct proc *p; p = td ? td->td_proc : NULL; if (error) { tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server, msg, error); } else { tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg); } return (0); } static void nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg, int error, int flags) { if (nmp == NULL) return; mtx_lock(&nmp->nm_mtx); if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) { nmp->nm_state |= NFSSTA_TIMEO; mtx_unlock(&nmp->nm_mtx); vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, VQ_NOTRESP, 0); } else mtx_unlock(&nmp->nm_mtx); mtx_lock(&nmp->nm_mtx); if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) { nmp->nm_state |= NFSSTA_LOCKTIMEO; mtx_unlock(&nmp->nm_mtx); vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, VQ_NOTRESPLOCK, 0); } else mtx_unlock(&nmp->nm_mtx); nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error); } static void nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg, int flags, int tprintfmsg) { if (nmp == NULL) return; if (tprintfmsg) { nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0); } mtx_lock(&nmp->nm_mtx); if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) { nmp->nm_state &= ~NFSSTA_TIMEO; mtx_unlock(&nmp->nm_mtx); vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, VQ_NOTRESP, 1); } else mtx_unlock(&nmp->nm_mtx); mtx_lock(&nmp->nm_mtx); if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) { nmp->nm_state &= ~NFSSTA_LOCKTIMEO; mtx_unlock(&nmp->nm_mtx); vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid, VQ_NOTRESPLOCK, 1); } else mtx_unlock(&nmp->nm_mtx); } Index: head/sys/fs/nfs/nfs_commonsubs.c =================================================================== --- head/sys/fs/nfs/nfs_commonsubs.c (revision 321627) +++ head/sys/fs/nfs/nfs_commonsubs.c (revision 321628) @@ -1,4243 +1,4241 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); /* * These functions support the macros and help fiddle mbuf chains for * the nfs op functions. They do things like create the rpc header and * copy data between mbuf chains and uio lists. */ #ifndef APPLEKEXT #include "opt_inet6.h" #include #include /* * Data items converted to xdr at startup, since they are constant * This is kinda hokey, but may save a little time doing byte swaps */ u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; /* And other global data */ nfstype nfsv34_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON }; enum vtype newnv2tov_type[8] = { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON }; enum vtype nv34tov_type[8]={ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO }; struct timeval nfsboottime; /* Copy boottime once, so it never changes */ int nfscl_ticks; int nfsrv_useacl = 1; struct nfssockreq nfsrv_nfsuserdsock; int nfsrv_nfsuserd = 0; struct nfsreqhead nfsd_reqq; uid_t nfsrv_defaultuid = UID_NOBODY; gid_t nfsrv_defaultgid = GID_NOGROUP; int nfsrv_lease = NFSRV_LEASE; int ncl_mbuf_mlen = MLEN; int nfsd_enable_stringtouid = 0; static int nfs_enable_uidtostring = 0; NFSNAMEIDMUTEX; NFSSOCKMUTEX; extern int nfsrv_lughashsize; SYSCTL_DECL(_vfs_nfs); SYSCTL_INT(_vfs_nfs, OID_AUTO, enable_uidtostring, CTLFLAG_RW, &nfs_enable_uidtostring, 0, "Make nfs always send numeric owner_names"); /* * This array of structures indicates, for V4: * retfh - which of 3 types of calling args are used * 0 - doesn't change cfh or use a sfh * 1 - replaces cfh with a new one (unless it returns an error status) * 2 - uses cfh and sfh * needscfh - if the op wants a cfh and premtime * 0 - doesn't use a cfh * 1 - uses a cfh, but doesn't want pre-op attributes * 2 - uses a cfh and wants pre-op attributes * savereply - indicates a non-idempotent Op * 0 - not non-idempotent * 1 - non-idempotent * Ops that are ordered via seqid# are handled separately from these * non-idempotent Ops. * Define it here, since it is used by both the client and server. */ struct nfsv4_opflag nfsv4_opflag[NFSV41_NOPS] = { { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* undef */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* undef */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* undef */ { 0, 1, 0, 0, LK_SHARED, 1, 1 }, /* Access */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Close */ { 0, 2, 0, 1, LK_EXCLUSIVE, 1, 1 }, /* Commit */ { 1, 2, 1, 1, LK_EXCLUSIVE, 1, 1 }, /* Create */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Delegpurge */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Delegreturn */ { 0, 1, 0, 0, LK_SHARED, 1, 1 }, /* Getattr */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* GetFH */ { 2, 1, 1, 1, LK_EXCLUSIVE, 1, 1 }, /* Link */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Lock */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* LockT */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* LockU */ { 1, 2, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Lookup */ { 1, 2, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Lookupp */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* NVerify */ { 1, 1, 0, 1, LK_EXCLUSIVE, 1, 0 }, /* Open */ { 1, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* OpenAttr */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* OpenConfirm */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* OpenDowngrade */ { 1, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* PutFH */ { 1, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* PutPubFH */ { 1, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* PutRootFH */ { 0, 1, 0, 0, LK_SHARED, 1, 0 }, /* Read */ { 0, 1, 0, 0, LK_SHARED, 1, 1 }, /* Readdir */ { 0, 1, 0, 0, LK_SHARED, 1, 1 }, /* ReadLink */ { 0, 2, 1, 1, LK_EXCLUSIVE, 1, 1 }, /* Remove */ { 2, 1, 1, 1, LK_EXCLUSIVE, 1, 1 }, /* Rename */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Renew */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* RestoreFH */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* SaveFH */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* SecInfo */ { 0, 2, 1, 1, LK_EXCLUSIVE, 1, 0 }, /* Setattr */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* SetClientID */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* SetClientIDConfirm */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Verify */ { 0, 2, 1, 1, LK_EXCLUSIVE, 1, 0 }, /* Write */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* ReleaseLockOwner */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Backchannel Ctrl */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Bind Conn to Sess */ { 0, 0, 0, 0, LK_EXCLUSIVE, 0, 0 }, /* Exchange ID */ { 0, 0, 0, 0, LK_EXCLUSIVE, 0, 0 }, /* Create Session */ { 0, 0, 0, 0, LK_EXCLUSIVE, 0, 0 }, /* Destroy Session */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Free StateID */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Get Dir Deleg */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Get Device Info */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Get Device List */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Layout Commit */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Layout Get */ { 0, 1, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Layout Return */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Secinfo No name */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Sequence */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Set SSV */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Test StateID */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 1 }, /* Want Delegation */ { 0, 0, 0, 0, LK_EXCLUSIVE, 0, 0 }, /* Destroy ClientID */ { 0, 0, 0, 0, LK_EXCLUSIVE, 1, 0 }, /* Reclaim Complete */ }; #endif /* !APPLEKEXT */ static int ncl_mbuf_mhlen = MHLEN; static int nfsrv_usercnt = 0; static int nfsrv_dnsnamelen; static u_char *nfsrv_dnsname = NULL; static int nfsrv_usermax = 999999999; struct nfsrv_lughash { struct mtx mtx; struct nfsuserhashhead lughead; }; static struct nfsrv_lughash *nfsuserhash; static struct nfsrv_lughash *nfsusernamehash; static struct nfsrv_lughash *nfsgrouphash; static struct nfsrv_lughash *nfsgroupnamehash; /* * This static array indicates whether or not the RPC generates a large * reply. This is used by nfs_reply() to decide whether or not an mbuf * cluster should be allocated. (If a cluster is required by an RPC * marked 0 in this array, the code will still work, just not quite as * efficiently.) */ int nfs_bigreply[NFSV41_NPROCS] = { 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0 }; /* local functions */ static int nfsrv_skipace(struct nfsrv_descript *nd, int *acesizep); static void nfsv4_wanted(struct nfsv4lock *lp); static int nfsrv_cmpmixedcase(u_char *cp, u_char *cp2, int len); static int nfsrv_getuser(int procnum, uid_t uid, gid_t gid, char *name, NFSPROC_T *p); static void nfsrv_removeuser(struct nfsusrgrp *usrp, int isuser); static int nfsrv_getrefstr(struct nfsrv_descript *, u_char **, u_char **, int *, int *); static void nfsrv_refstrbigenough(int, u_char **, u_char **, int *); #ifndef APPLE /* * copies mbuf chain to the uio scatter/gather list */ int nfsm_mbufuio(struct nfsrv_descript *nd, struct uio *uiop, int siz) { char *mbufcp, *uiocp; int xfer, left, len; mbuf_t mp; long uiosiz, rem; int error = 0; mp = nd->nd_md; mbufcp = nd->nd_dpos; len = NFSMTOD(mp, caddr_t) + mbuf_len(mp) - mbufcp; rem = NFSM_RNDUP(siz) - siz; while (siz > 0) { if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL) { error = EBADRPC; goto out; } left = uiop->uio_iov->iov_len; uiocp = uiop->uio_iov->iov_base; if (left > siz) left = siz; uiosiz = left; while (left > 0) { while (len == 0) { mp = mbuf_next(mp); if (mp == NULL) { error = EBADRPC; goto out; } mbufcp = NFSMTOD(mp, caddr_t); len = mbuf_len(mp); KASSERT(len >= 0, ("len %d, corrupted mbuf?", len)); } xfer = (left > len) ? len : left; #ifdef notdef /* Not Yet.. */ if (uiop->uio_iov->iov_op != NULL) (*(uiop->uio_iov->iov_op)) (mbufcp, uiocp, xfer); else #endif if (uiop->uio_segflg == UIO_SYSSPACE) NFSBCOPY(mbufcp, uiocp, xfer); else copyout(mbufcp, CAST_USER_ADDR_T(uiocp), xfer); left -= xfer; len -= xfer; mbufcp += xfer; uiocp += xfer; uiop->uio_offset += xfer; uiop->uio_resid -= xfer; } if (uiop->uio_iov->iov_len <= siz) { uiop->uio_iovcnt--; uiop->uio_iov++; } else { uiop->uio_iov->iov_base = (void *) ((char *)uiop->uio_iov->iov_base + uiosiz); uiop->uio_iov->iov_len -= uiosiz; } siz -= uiosiz; } nd->nd_dpos = mbufcp; nd->nd_md = mp; if (rem > 0) { if (len < rem) error = nfsm_advance(nd, rem, len); else nd->nd_dpos += rem; } out: NFSEXITCODE2(error, nd); return (error); } #endif /* !APPLE */ /* * Help break down an mbuf chain by setting the first siz bytes contiguous * pointed to by returned val. * This is used by the macro NFSM_DISSECT for tough * cases. */ APPLESTATIC void * nfsm_dissct(struct nfsrv_descript *nd, int siz, int how) { mbuf_t mp2; int siz2, xfer; caddr_t p; int left; caddr_t retp; retp = NULL; left = NFSMTOD(nd->nd_md, caddr_t) + mbuf_len(nd->nd_md) - nd->nd_dpos; while (left == 0) { nd->nd_md = mbuf_next(nd->nd_md); if (nd->nd_md == NULL) return (retp); left = mbuf_len(nd->nd_md); nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t); } if (left >= siz) { retp = nd->nd_dpos; nd->nd_dpos += siz; } else if (mbuf_next(nd->nd_md) == NULL) { return (retp); } else if (siz > ncl_mbuf_mhlen) { panic("nfs S too big"); } else { MGET(mp2, MT_DATA, how); if (mp2 == NULL) return (NULL); mbuf_setnext(mp2, mbuf_next(nd->nd_md)); mbuf_setnext(nd->nd_md, mp2); mbuf_setlen(nd->nd_md, mbuf_len(nd->nd_md) - left); nd->nd_md = mp2; retp = p = NFSMTOD(mp2, caddr_t); NFSBCOPY(nd->nd_dpos, p, left); /* Copy what was left */ siz2 = siz - left; p += left; mp2 = mbuf_next(mp2); /* Loop around copying up the siz2 bytes */ while (siz2 > 0) { if (mp2 == NULL) return (NULL); xfer = (siz2 > mbuf_len(mp2)) ? mbuf_len(mp2) : siz2; if (xfer > 0) { NFSBCOPY(NFSMTOD(mp2, caddr_t), p, xfer); NFSM_DATAP(mp2, xfer); mbuf_setlen(mp2, mbuf_len(mp2) - xfer); p += xfer; siz2 -= xfer; } if (siz2 > 0) mp2 = mbuf_next(mp2); } mbuf_setlen(nd->nd_md, siz); nd->nd_md = mp2; nd->nd_dpos = NFSMTOD(mp2, caddr_t); } return (retp); } /* * Advance the position in the mbuf chain. * If offs == 0, this is a no-op, but it is simpler to just return from * here than check for offs > 0 for all calls to nfsm_advance. * If left == -1, it should be calculated here. */ APPLESTATIC int nfsm_advance(struct nfsrv_descript *nd, int offs, int left) { int error = 0; if (offs == 0) goto out; /* * A negative offs should be considered a serious problem. */ if (offs < 0) panic("nfsrv_advance"); /* * If left == -1, calculate it here. */ if (left == -1) left = NFSMTOD(nd->nd_md, caddr_t) + mbuf_len(nd->nd_md) - nd->nd_dpos; /* * Loop around, advancing over the mbuf data. */ while (offs > left) { offs -= left; nd->nd_md = mbuf_next(nd->nd_md); if (nd->nd_md == NULL) { error = EBADRPC; goto out; } left = mbuf_len(nd->nd_md); nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t); } nd->nd_dpos += offs; out: NFSEXITCODE(error); return (error); } /* * Copy a string into mbuf(s). * Return the number of bytes output, including XDR overheads. */ APPLESTATIC int nfsm_strtom(struct nfsrv_descript *nd, const char *cp, int siz) { mbuf_t m2; int xfer, left; mbuf_t m1; int rem, bytesize; u_int32_t *tl; char *cp2; NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(siz); rem = NFSM_RNDUP(siz) - siz; bytesize = NFSX_UNSIGNED + siz + rem; m2 = nd->nd_mb; cp2 = nd->nd_bpos; left = M_TRAILINGSPACE(m2); /* * Loop around copying the string to mbuf(s). */ while (siz > 0) { if (left == 0) { if (siz > ncl_mbuf_mlen) NFSMCLGET(m1, M_WAITOK); else NFSMGET(m1); mbuf_setlen(m1, 0); mbuf_setnext(m2, m1); m2 = m1; cp2 = NFSMTOD(m2, caddr_t); left = M_TRAILINGSPACE(m2); } if (left >= siz) xfer = siz; else xfer = left; NFSBCOPY(cp, cp2, xfer); cp += xfer; mbuf_setlen(m2, mbuf_len(m2) + xfer); siz -= xfer; left -= xfer; if (siz == 0 && rem) { if (left < rem) panic("nfsm_strtom"); NFSBZERO(cp2 + xfer, rem); mbuf_setlen(m2, mbuf_len(m2) + rem); } } nd->nd_mb = m2; nd->nd_bpos = NFSMTOD(m2, caddr_t) + mbuf_len(m2); return (bytesize); } /* * Called once to initialize data structures... */ APPLESTATIC void newnfs_init(void) { static int nfs_inited = 0; if (nfs_inited) return; nfs_inited = 1; newnfs_true = txdr_unsigned(TRUE); newnfs_false = txdr_unsigned(FALSE); newnfs_xdrneg1 = txdr_unsigned(-1); nfscl_ticks = (hz * NFS_TICKINTVL + 500) / 1000; if (nfscl_ticks < 1) nfscl_ticks = 1; NFSSETBOOTTIME(nfsboottime); /* * Initialize reply list and start timer */ TAILQ_INIT(&nfsd_reqq); NFS_TIMERINIT; } /* * Put a file handle in an mbuf list. * If the size argument == 0, just use the default size. * set_true == 1 if there should be an newnfs_true prepended on the file handle. * Return the number of bytes output, including XDR overhead. */ APPLESTATIC int nfsm_fhtom(struct nfsrv_descript *nd, u_int8_t *fhp, int size, int set_true) { u_int32_t *tl; u_int8_t *cp; int fullsiz, rem, bytesize = 0; if (size == 0) size = NFSX_MYFH; switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { case ND_NFSV2: if (size > NFSX_V2FH) panic("fh size > NFSX_V2FH for NFSv2"); NFSM_BUILD(cp, u_int8_t *, NFSX_V2FH); NFSBCOPY(fhp, cp, size); if (size < NFSX_V2FH) NFSBZERO(cp + size, NFSX_V2FH - size); bytesize = NFSX_V2FH; break; case ND_NFSV3: case ND_NFSV4: fullsiz = NFSM_RNDUP(size); rem = fullsiz - size; if (set_true) { bytesize = 2 * NFSX_UNSIGNED + fullsiz; NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_true; } else { bytesize = NFSX_UNSIGNED + fullsiz; } (void) nfsm_strtom(nd, fhp, size); break; } return (bytesize); } /* * This function compares two net addresses by family and returns TRUE * if they are the same host. * If there is any doubt, return FALSE. * The AF_INET family is handled as a special case so that address mbufs * don't need to be saved to store "struct in_addr", which is only 4 bytes. */ APPLESTATIC int nfsaddr_match(int family, union nethostaddr *haddr, NFSSOCKADDR_T nam) { struct sockaddr_in *inetaddr; switch (family) { case AF_INET: inetaddr = NFSSOCKADDR(nam, struct sockaddr_in *); if (inetaddr->sin_family == AF_INET && inetaddr->sin_addr.s_addr == haddr->had_inet.s_addr) return (1); break; #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *inetaddr6; inetaddr6 = NFSSOCKADDR(nam, struct sockaddr_in6 *); /* XXX - should test sin6_scope_id ? */ if (inetaddr6->sin6_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(&inetaddr6->sin6_addr, &haddr->had_inet6)) return (1); } break; #endif } return (0); } /* * Similar to the above, but takes to NFSSOCKADDR_T args. */ APPLESTATIC int nfsaddr2_match(NFSSOCKADDR_T nam1, NFSSOCKADDR_T nam2) { struct sockaddr_in *addr1, *addr2; struct sockaddr *inaddr; inaddr = NFSSOCKADDR(nam1, struct sockaddr *); switch (inaddr->sa_family) { case AF_INET: addr1 = NFSSOCKADDR(nam1, struct sockaddr_in *); addr2 = NFSSOCKADDR(nam2, struct sockaddr_in *); if (addr2->sin_family == AF_INET && addr1->sin_addr.s_addr == addr2->sin_addr.s_addr) return (1); break; #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *inet6addr1, *inet6addr2; inet6addr1 = NFSSOCKADDR(nam1, struct sockaddr_in6 *); inet6addr2 = NFSSOCKADDR(nam2, struct sockaddr_in6 *); /* XXX - should test sin6_scope_id ? */ if (inet6addr2->sin6_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(&inet6addr1->sin6_addr, &inet6addr2->sin6_addr)) return (1); } break; #endif } return (0); } /* * Trim the stuff already dissected off the mbuf list. */ APPLESTATIC void newnfs_trimleading(nd) struct nfsrv_descript *nd; { mbuf_t m, n; int offs; /* * First, free up leading mbufs. */ if (nd->nd_mrep != nd->nd_md) { m = nd->nd_mrep; while (mbuf_next(m) != nd->nd_md) { if (mbuf_next(m) == NULL) panic("nfsm trim leading"); m = mbuf_next(m); } mbuf_setnext(m, NULL); mbuf_freem(nd->nd_mrep); } m = nd->nd_md; /* * Now, adjust this mbuf, based on nd_dpos. */ offs = nd->nd_dpos - NFSMTOD(m, caddr_t); if (offs == mbuf_len(m)) { n = m; m = mbuf_next(m); if (m == NULL) panic("nfsm trim leading2"); mbuf_setnext(n, NULL); mbuf_freem(n); } else if (offs > 0) { mbuf_setlen(m, mbuf_len(m) - offs); NFSM_DATAP(m, offs); } else if (offs < 0) panic("nfsm trimleading offs"); nd->nd_mrep = m; nd->nd_md = m; nd->nd_dpos = NFSMTOD(m, caddr_t); } /* * Trim trailing data off the mbuf list being built. */ APPLESTATIC void newnfs_trimtrailing(nd, mb, bpos) struct nfsrv_descript *nd; mbuf_t mb; caddr_t bpos; { if (mbuf_next(mb)) { mbuf_freem(mbuf_next(mb)); mbuf_setnext(mb, NULL); } mbuf_setlen(mb, bpos - NFSMTOD(mb, caddr_t)); nd->nd_mb = mb; nd->nd_bpos = bpos; } /* * Dissect a file handle on the client. */ APPLESTATIC int nfsm_getfh(struct nfsrv_descript *nd, struct nfsfh **nfhpp) { u_int32_t *tl; struct nfsfh *nfhp; int error, len; *nfhpp = NULL; if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if ((len = fxdr_unsigned(int, *tl)) <= 0 || len > NFSX_FHMAX) { error = EBADRPC; goto nfsmout; } } else len = NFSX_V2FH; MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + len, M_NFSFH, M_WAITOK); error = nfsrv_mtostr(nd, nfhp->nfh_fh, len); if (error) { FREE((caddr_t)nfhp, M_NFSFH); goto nfsmout; } nfhp->nfh_len = len; *nfhpp = nfhp; nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Break down the nfsv4 acl. * If the aclp == NULL or won't fit in an acl, just discard the acl info. */ APPLESTATIC int nfsrv_dissectacl(struct nfsrv_descript *nd, NFSACL_T *aclp, int *aclerrp, int *aclsizep, __unused NFSPROC_T *p) { u_int32_t *tl; int i, aclsize; int acecnt, error = 0, aceerr = 0, acesize; *aclerrp = 0; if (aclp) aclp->acl_cnt = 0; /* * Parse out the ace entries and expect them to conform to * what can be supported by R/W/X bits. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); aclsize = NFSX_UNSIGNED; acecnt = fxdr_unsigned(int, *tl); if (acecnt > ACL_MAX_ENTRIES) aceerr = NFSERR_ATTRNOTSUPP; if (nfsrv_useacl == 0) aceerr = NFSERR_ATTRNOTSUPP; for (i = 0; i < acecnt; i++) { if (aclp && !aceerr) error = nfsrv_dissectace(nd, &aclp->acl_entry[i], &aceerr, &acesize, p); else error = nfsrv_skipace(nd, &acesize); if (error) goto nfsmout; aclsize += acesize; } if (aclp && !aceerr) aclp->acl_cnt = acecnt; if (aceerr) *aclerrp = aceerr; if (aclsizep) *aclsizep = aclsize; nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Skip over an NFSv4 ace entry. Just dissect the xdr and discard it. */ static int nfsrv_skipace(struct nfsrv_descript *nd, int *acesizep) { u_int32_t *tl; int error, len = 0; NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED); len = fxdr_unsigned(int, *(tl + 3)); error = nfsm_advance(nd, NFSM_RNDUP(len), -1); nfsmout: *acesizep = NFSM_RNDUP(len) + (4 * NFSX_UNSIGNED); NFSEXITCODE2(error, nd); return (error); } /* * Get attribute bits from an mbuf list. * Returns EBADRPC for a parsing error, 0 otherwise. * If the clearinvalid flag is set, clear the bits not supported. */ APPLESTATIC int nfsrv_getattrbits(struct nfsrv_descript *nd, nfsattrbit_t *attrbitp, int *cntp, int *retnotsupp) { u_int32_t *tl; int cnt, i, outcnt; int error = 0; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); cnt = fxdr_unsigned(int, *tl); if (cnt < 0) { error = NFSERR_BADXDR; goto nfsmout; } if (cnt > NFSATTRBIT_MAXWORDS) outcnt = NFSATTRBIT_MAXWORDS; else outcnt = cnt; NFSZERO_ATTRBIT(attrbitp); if (outcnt > 0) { NFSM_DISSECT(tl, u_int32_t *, outcnt * NFSX_UNSIGNED); for (i = 0; i < outcnt; i++) attrbitp->bits[i] = fxdr_unsigned(u_int32_t, *tl++); } for (i = 0; i < (cnt - outcnt); i++) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (retnotsupp != NULL && *tl != 0) *retnotsupp = NFSERR_ATTRNOTSUPP; } if (cntp) *cntp = NFSX_UNSIGNED + (cnt * NFSX_UNSIGNED); nfsmout: NFSEXITCODE2(error, nd); return (error); } /* * Get the attributes for V4. * If the compare flag is true, test for any attribute changes, * otherwise return the attribute values. * These attributes cover fields in "struct vattr", "struct statfs", * "struct nfsfsinfo", the file handle and the lease duration. * The value of retcmpp is set to 1 if all attributes are the same, * and 0 otherwise. * Returns EBADRPC if it can't be parsed, 0 otherwise. */ APPLESTATIC int nfsv4_loadattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nap, struct nfsfh **nfhpp, fhandle_t *fhp, int fhsize, struct nfsv3_pathconf *pc, struct statfs *sbp, struct nfsstatfs *sfp, struct nfsfsinfo *fsp, NFSACL_T *aclp, int compare, int *retcmpp, u_int32_t *leasep, u_int32_t *rderrp, NFSPROC_T *p, struct ucred *cred) { u_int32_t *tl; int i = 0, j, k, l = 0, m, bitpos, attrsum = 0; int error, tfhsize, aceerr, attrsize, cnt, retnotsup; u_char *cp, *cp2, namestr[NFSV4_SMALLSTR + 1]; nfsattrbit_t attrbits, retattrbits, checkattrbits; struct nfsfh *tnfhp; struct nfsreferral *refp; u_quad_t tquad; nfsquad_t tnfsquad; struct timespec temptime; uid_t uid; gid_t gid; u_int32_t freenum = 0, tuint; u_int64_t uquad = 0, thyp, thyp2; #ifdef QUOTA struct dqblk dqb; uid_t savuid; #endif CTASSERT(sizeof(ino_t) == sizeof(uint64_t)); if (compare) { retnotsup = 0; error = nfsrv_getattrbits(nd, &attrbits, NULL, &retnotsup); } else { error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); } if (error) goto nfsmout; if (compare) { *retcmpp = retnotsup; } else { /* * Just set default values to some of the important ones. */ if (nap != NULL) { nap->na_type = VREG; nap->na_mode = 0; nap->na_rdev = (NFSDEV_T)0; nap->na_mtime.tv_sec = 0; nap->na_mtime.tv_nsec = 0; nap->na_gen = 0; nap->na_flags = 0; nap->na_blocksize = NFS_FABLKSIZE; } if (sbp != NULL) { sbp->f_bsize = NFS_FABLKSIZE; sbp->f_blocks = 0; sbp->f_bfree = 0; sbp->f_bavail = 0; sbp->f_files = 0; sbp->f_ffree = 0; } if (fsp != NULL) { fsp->fs_rtmax = 8192; fsp->fs_rtpref = 8192; fsp->fs_maxname = NFS_MAXNAMLEN; fsp->fs_wtmax = 8192; fsp->fs_wtpref = 8192; fsp->fs_wtmult = NFS_FABLKSIZE; fsp->fs_dtpref = 8192; fsp->fs_maxfilesize = 0xffffffffffffffffull; fsp->fs_timedelta.tv_sec = 0; fsp->fs_timedelta.tv_nsec = 1; fsp->fs_properties = (NFSV3_FSFLINK | NFSV3_FSFSYMLINK | NFSV3_FSFHOMOGENEOUS | NFSV3_FSFCANSETTIME); } if (pc != NULL) { pc->pc_linkmax = LINK_MAX; pc->pc_namemax = NAME_MAX; pc->pc_notrunc = 0; pc->pc_chownrestricted = 0; pc->pc_caseinsensitive = 0; pc->pc_casepreserving = 1; } if (sfp != NULL) { sfp->sf_ffiles = UINT64_MAX; sfp->sf_tfiles = UINT64_MAX; sfp->sf_afiles = UINT64_MAX; sfp->sf_fbytes = UINT64_MAX; sfp->sf_tbytes = UINT64_MAX; sfp->sf_abytes = UINT64_MAX; } } /* * Loop around getting the attributes. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsize = fxdr_unsigned(int, *tl); for (bitpos = 0; bitpos < NFSATTRBIT_MAX; bitpos++) { if (attrsum > attrsize) { error = NFSERR_BADXDR; goto nfsmout; } if (NFSISSET_ATTRBIT(&attrbits, bitpos)) switch (bitpos) { case NFSATTRBIT_SUPPORTEDATTRS: retnotsup = 0; if (compare || nap == NULL) error = nfsrv_getattrbits(nd, &retattrbits, &cnt, &retnotsup); else error = nfsrv_getattrbits(nd, &nap->na_suppattr, &cnt, &retnotsup); if (error) goto nfsmout; if (compare && !(*retcmpp)) { NFSSETSUPP_ATTRBIT(&checkattrbits); /* Some filesystem do not support NFSv4ACL */ if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) { NFSCLRBIT_ATTRBIT(&checkattrbits, NFSATTRBIT_ACL); NFSCLRBIT_ATTRBIT(&checkattrbits, NFSATTRBIT_ACLSUPPORT); } if (!NFSEQUAL_ATTRBIT(&retattrbits, &checkattrbits) || retnotsup) *retcmpp = NFSERR_NOTSAME; } attrsum += cnt; break; case NFSATTRBIT_TYPE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (nap->na_type != nfsv34tov_type(*tl)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_type = nfsv34tov_type(*tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_FHEXPIRETYPE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) { if (fxdr_unsigned(int, *tl) != NFSV4FHTYPE_PERSISTENT) *retcmpp = NFSERR_NOTSAME; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_CHANGE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp)) { if (nap->na_filerev != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_filerev = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_SIZE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp)) { if (nap->na_size != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_size = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_LINKSUPPORT: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (fsp->fs_properties & NFSV3_FSFLINK) { if (*tl == newnfs_false) *retcmpp = NFSERR_NOTSAME; } else { if (*tl == newnfs_true) *retcmpp = NFSERR_NOTSAME; } } } else if (fsp != NULL) { if (*tl == newnfs_true) fsp->fs_properties |= NFSV3_FSFLINK; else fsp->fs_properties &= ~NFSV3_FSFLINK; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_SYMLINKSUPPORT: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (fsp->fs_properties & NFSV3_FSFSYMLINK) { if (*tl == newnfs_false) *retcmpp = NFSERR_NOTSAME; } else { if (*tl == newnfs_true) *retcmpp = NFSERR_NOTSAME; } } } else if (fsp != NULL) { if (*tl == newnfs_true) fsp->fs_properties |= NFSV3_FSFSYMLINK; else fsp->fs_properties &= ~NFSV3_FSFSYMLINK; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_NAMEDATTR: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) { if (*tl != newnfs_false) *retcmpp = NFSERR_NOTSAME; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_FSID: NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED); thyp = fxdr_hyper(tl); tl += 2; thyp2 = fxdr_hyper(tl); if (compare) { if (*retcmpp == 0) { if (thyp != (u_int64_t) vfs_statfs(vnode_mount(vp))->f_fsid.val[0] || thyp2 != (u_int64_t) vfs_statfs(vnode_mount(vp))->f_fsid.val[1]) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_filesid[0] = thyp; nap->na_filesid[1] = thyp2; } attrsum += (4 * NFSX_UNSIGNED); break; case NFSATTRBIT_UNIQUEHANDLES: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) { if (*tl != newnfs_true) *retcmpp = NFSERR_NOTSAME; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_LEASETIME: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (fxdr_unsigned(int, *tl) != nfsrv_lease && !(*retcmpp)) *retcmpp = NFSERR_NOTSAME; } else if (leasep != NULL) { *leasep = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_RDATTRERROR: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) *retcmpp = NFSERR_INVAL; } else if (rderrp != NULL) { *rderrp = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_ACL: if (compare) { if (!(*retcmpp)) { if (nfsrv_useacl && nfs_supportsnfsv4acls(vp)) { NFSACL_T *naclp; naclp = acl_alloc(M_WAITOK); error = nfsrv_dissectacl(nd, naclp, &aceerr, &cnt, p); if (error) { acl_free(naclp); goto nfsmout; } if (aceerr || aclp == NULL || nfsrv_compareacl(aclp, naclp)) *retcmpp = NFSERR_NOTSAME; acl_free(naclp); } else { error = nfsrv_dissectacl(nd, NULL, &aceerr, &cnt, p); *retcmpp = NFSERR_ATTRNOTSUPP; } } } else { if (vp != NULL && aclp != NULL) error = nfsrv_dissectacl(nd, aclp, &aceerr, &cnt, p); else error = nfsrv_dissectacl(nd, NULL, &aceerr, &cnt, p); if (error) goto nfsmout; } attrsum += cnt; break; case NFSATTRBIT_ACLSUPPORT: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) { if (nfsrv_useacl && nfs_supportsnfsv4acls(vp)) { if (fxdr_unsigned(u_int32_t, *tl) != NFSV4ACE_SUPTYPES) *retcmpp = NFSERR_NOTSAME; } else { *retcmpp = NFSERR_ATTRNOTSUPP; } } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_ARCHIVE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_CANSETTIME: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (fsp->fs_properties & NFSV3_FSFCANSETTIME) { if (*tl == newnfs_false) *retcmpp = NFSERR_NOTSAME; } else { if (*tl == newnfs_true) *retcmpp = NFSERR_NOTSAME; } } } else if (fsp != NULL) { if (*tl == newnfs_true) fsp->fs_properties |= NFSV3_FSFCANSETTIME; else fsp->fs_properties &= ~NFSV3_FSFCANSETTIME; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_CASEINSENSITIVE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (*tl != newnfs_false) *retcmpp = NFSERR_NOTSAME; } } else if (pc != NULL) { pc->pc_caseinsensitive = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_CASEPRESERVING: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (*tl != newnfs_true) *retcmpp = NFSERR_NOTSAME; } } else if (pc != NULL) { pc->pc_casepreserving = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_CHOWNRESTRICTED: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (*tl != newnfs_true) *retcmpp = NFSERR_NOTSAME; } } else if (pc != NULL) { pc->pc_chownrestricted = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_FILEHANDLE: error = nfsm_getfh(nd, &tnfhp); if (error) goto nfsmout; tfhsize = tnfhp->nfh_len; if (compare) { if (!(*retcmpp) && !NFSRV_CMPFH(tnfhp->nfh_fh, tfhsize, fhp, fhsize)) *retcmpp = NFSERR_NOTSAME; FREE((caddr_t)tnfhp, M_NFSFH); } else if (nfhpp != NULL) { *nfhpp = tnfhp; } else { FREE((caddr_t)tnfhp, M_NFSFH); } attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(tfhsize)); break; case NFSATTRBIT_FILEID: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); thyp = fxdr_hyper(tl); if (compare) { if (!(*retcmpp)) { if (nap->na_fileid != thyp) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) nap->na_fileid = thyp; attrsum += NFSX_HYPER; break; case NFSATTRBIT_FILESAVAIL: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp) && sfp->sf_afiles != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } else if (sfp != NULL) { sfp->sf_afiles = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_FILESFREE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp) && sfp->sf_ffiles != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } else if (sfp != NULL) { sfp->sf_ffiles = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_FILESTOTAL: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp) && sfp->sf_tfiles != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } else if (sfp != NULL) { sfp->sf_tfiles = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_FSLOCATIONS: error = nfsrv_getrefstr(nd, &cp, &cp2, &l, &m); if (error) goto nfsmout; attrsum += l; if (compare && !(*retcmpp)) { refp = nfsv4root_getreferral(vp, NULL, 0); if (refp != NULL) { if (cp == NULL || cp2 == NULL || strcmp(cp, "/") || strcmp(cp2, refp->nfr_srvlist)) *retcmpp = NFSERR_NOTSAME; } else if (m == 0) { *retcmpp = NFSERR_NOTSAME; } } if (cp != NULL) free(cp, M_NFSSTRING); if (cp2 != NULL) free(cp2, M_NFSSTRING); break; case NFSATTRBIT_HIDDEN: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_HOMOGENEOUS: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (fsp->fs_properties & NFSV3_FSFHOMOGENEOUS) { if (*tl == newnfs_false) *retcmpp = NFSERR_NOTSAME; } else { if (*tl == newnfs_true) *retcmpp = NFSERR_NOTSAME; } } } else if (fsp != NULL) { if (*tl == newnfs_true) fsp->fs_properties |= NFSV3_FSFHOMOGENEOUS; else fsp->fs_properties &= ~NFSV3_FSFHOMOGENEOUS; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_MAXFILESIZE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); tnfsquad.qval = fxdr_hyper(tl); if (compare) { if (!(*retcmpp)) { tquad = NFSRV_MAXFILESIZE; if (tquad != tnfsquad.qval) *retcmpp = NFSERR_NOTSAME; } } else if (fsp != NULL) { fsp->fs_maxfilesize = tnfsquad.qval; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_MAXLINK: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (fxdr_unsigned(int, *tl) != LINK_MAX) *retcmpp = NFSERR_NOTSAME; } } else if (pc != NULL) { pc->pc_linkmax = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_MAXNAME: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (fsp->fs_maxname != fxdr_unsigned(u_int32_t, *tl)) *retcmpp = NFSERR_NOTSAME; } } else { tuint = fxdr_unsigned(u_int32_t, *tl); /* * Some Linux NFSv4 servers report this * as 0 or 4billion, so I'll set it to * NFS_MAXNAMLEN. If a server actually creates * a name longer than NFS_MAXNAMLEN, it will * get an error back. */ if (tuint == 0 || tuint > NFS_MAXNAMLEN) tuint = NFS_MAXNAMLEN; if (fsp != NULL) fsp->fs_maxname = tuint; if (pc != NULL) pc->pc_namemax = tuint; } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_MAXREAD: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp)) { if (fsp->fs_rtmax != fxdr_unsigned(u_int32_t, *(tl + 1)) || *tl != 0) *retcmpp = NFSERR_NOTSAME; } } else if (fsp != NULL) { fsp->fs_rtmax = fxdr_unsigned(u_int32_t, *++tl); fsp->fs_rtpref = fsp->fs_rtmax; fsp->fs_dtpref = fsp->fs_rtpref; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_MAXWRITE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp)) { if (fsp->fs_wtmax != fxdr_unsigned(u_int32_t, *(tl + 1)) || *tl != 0) *retcmpp = NFSERR_NOTSAME; } } else if (fsp != NULL) { fsp->fs_wtmax = fxdr_unsigned(int, *++tl); fsp->fs_wtpref = fsp->fs_wtmax; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_MIMETYPE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i)); error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; break; case NFSATTRBIT_MODE: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (nap->na_mode != nfstov_mode(*tl)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_mode = nfstov_mode(*tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_NOTRUNC: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare) { if (!(*retcmpp)) { if (*tl != newnfs_true) *retcmpp = NFSERR_NOTSAME; } } else if (pc != NULL) { pc->pc_notrunc = fxdr_unsigned(u_int32_t, *tl); } attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_NUMLINKS: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); tuint = fxdr_unsigned(u_int32_t, *tl); if (compare) { if (!(*retcmpp)) { if ((u_int32_t)nap->na_nlink != tuint) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_nlink = tuint; } 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; } attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 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 (compare) { if (!(*retcmpp)) { if (nfsv4_strtouid(nd, cp, j, &uid, p) || nap->na_uid != uid) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { if (nfsv4_strtouid(nd, cp, j, &uid, p)) nap->na_uid = nfsrv_defaultuid; else nap->na_uid = uid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); 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; } attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j)); 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 (compare) { if (!(*retcmpp)) { if (nfsv4_strtogid(nd, cp, j, &gid, p) || nap->na_gid != gid) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { if (nfsv4_strtogid(nd, cp, j, &gid, p)) nap->na_gid = nfsrv_defaultgid; else nap->na_gid = gid; } if (j > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); break; case NFSATTRBIT_QUOTAHARD: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (sbp != NULL) { if (priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0)) freenum = sbp->f_bfree; else freenum = sbp->f_bavail; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(vnode_mount(vp),QCMD(Q_GETQUOTA, USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = min(dqb.dqb_bhardlimit, freenum); p->p_cred->p_ruid = savuid; #endif /* QUOTA */ uquad = (u_int64_t)freenum; NFSQUOTABLKTOBYTE(uquad, sbp->f_bsize); } if (compare && !(*retcmpp)) { if (uquad != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_QUOTASOFT: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (sbp != NULL) { if (priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0)) freenum = sbp->f_bfree; else freenum = sbp->f_bavail; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(vnode_mount(vp),QCMD(Q_GETQUOTA, USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = min(dqb.dqb_bsoftlimit, freenum); p->p_cred->p_ruid = savuid; #endif /* QUOTA */ uquad = (u_int64_t)freenum; NFSQUOTABLKTOBYTE(uquad, sbp->f_bsize); } if (compare && !(*retcmpp)) { if (uquad != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_QUOTAUSED: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (sbp != NULL) { freenum = 0; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(vnode_mount(vp),QCMD(Q_GETQUOTA, USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = dqb.dqb_curblocks; p->p_cred->p_ruid = savuid; #endif /* QUOTA */ uquad = (u_int64_t)freenum; NFSQUOTABLKTOBYTE(uquad, sbp->f_bsize); } if (compare && !(*retcmpp)) { if (uquad != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_RAWDEV: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4SPECDATA); j = fxdr_unsigned(int, *tl++); k = fxdr_unsigned(int, *tl); if (compare) { if (!(*retcmpp)) { if (nap->na_rdev != NFSMAKEDEV(j, k)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_rdev = NFSMAKEDEV(j, k); } attrsum += NFSX_V4SPECDATA; break; case NFSATTRBIT_SPACEAVAIL: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp) && sfp->sf_abytes != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } else if (sfp != NULL) { sfp->sf_abytes = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_SPACEFREE: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp) && sfp->sf_fbytes != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } else if (sfp != NULL) { sfp->sf_fbytes = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_SPACETOTAL: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); if (compare) { if (!(*retcmpp) && sfp->sf_tbytes != fxdr_hyper(tl)) *retcmpp = NFSERR_NOTSAME; } else if (sfp != NULL) { sfp->sf_tbytes = fxdr_hyper(tl); } attrsum += NFSX_HYPER; break; case NFSATTRBIT_SPACEUSED: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); thyp = fxdr_hyper(tl); if (compare) { if (!(*retcmpp)) { if ((u_int64_t)nap->na_bytes != thyp) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_bytes = thyp; } attrsum += NFSX_HYPER; break; case NFSATTRBIT_SYSTEM: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; attrsum += NFSX_UNSIGNED; break; case NFSATTRBIT_TIMEACCESS: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &temptime); if (compare) { if (!(*retcmpp)) { if (!NFS_CMPTIME(temptime, nap->na_atime)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_atime = temptime; } attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEACCESSSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; i = fxdr_unsigned(int, *tl); if (i == NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); attrsum += NFSX_V4TIME; } if (compare && !(*retcmpp)) *retcmpp = NFSERR_INVAL; break; case NFSATTRBIT_TIMEBACKUP: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMECREATE: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEDELTA: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); if (fsp != NULL) { if (compare) { if (!(*retcmpp)) { if ((u_int32_t)fsp->fs_timedelta.tv_sec != fxdr_unsigned(u_int32_t, *(tl + 1)) || (u_int32_t)fsp->fs_timedelta.tv_nsec != (fxdr_unsigned(u_int32_t, *(tl + 2)) % 1000000000) || *tl != 0) *retcmpp = NFSERR_NOTSAME; } } else { fxdr_nfsv4time(tl, &fsp->fs_timedelta); } } attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMETADATA: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &temptime); if (compare) { if (!(*retcmpp)) { if (!NFS_CMPTIME(temptime, nap->na_ctime)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_ctime = temptime; } attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFY: NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); fxdr_nfsv4time(tl, &temptime); if (compare) { if (!(*retcmpp)) { if (!NFS_CMPTIME(temptime, nap->na_mtime)) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) { nap->na_mtime = temptime; } attrsum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFYSET: NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); attrsum += NFSX_UNSIGNED; i = fxdr_unsigned(int, *tl); if (i == NFSV4SATTRTIME_TOCLIENT) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME); attrsum += NFSX_V4TIME; } if (compare && !(*retcmpp)) *retcmpp = NFSERR_INVAL; break; case NFSATTRBIT_MOUNTEDONFILEID: NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); thyp = fxdr_hyper(tl); if (compare) { if (!(*retcmpp)) { if (!vp || !nfsrv_atroot(vp, &thyp2)) thyp2 = nap->na_fileid; if (thyp2 != thyp) *retcmpp = NFSERR_NOTSAME; } } else if (nap != NULL) nap->na_mntonfileno = thyp; attrsum += NFSX_HYPER; break; case NFSATTRBIT_SUPPATTREXCLCREAT: retnotsup = 0; error = nfsrv_getattrbits(nd, &retattrbits, &cnt, &retnotsup); if (error) goto nfsmout; if (compare && !(*retcmpp)) { NFSSETSUPP_ATTRBIT(&checkattrbits); NFSCLRNOTSETABLE_ATTRBIT(&checkattrbits); NFSCLRBIT_ATTRBIT(&checkattrbits, NFSATTRBIT_TIMEACCESSSET); if (!NFSEQUAL_ATTRBIT(&retattrbits, &checkattrbits) || retnotsup) *retcmpp = NFSERR_NOTSAME; } attrsum += cnt; break; default: printf("EEK! nfsv4_loadattr unknown attr=%d\n", bitpos); if (compare && !(*retcmpp)) *retcmpp = NFSERR_ATTRNOTSUPP; /* * and get out of the loop, since we can't parse * the unknown attrbute data. */ 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); } /* * Implement sleep locks for newnfs. The nfslock_usecnt allows for a * shared lock and the NFSXXX_LOCK flag permits an exclusive lock. * The first argument is a pointer to an nfsv4lock structure. * The second argument is 1 iff a blocking lock is wanted. * If this argument is 0, the call waits until no thread either wants nor * holds an exclusive lock. * It returns 1 if the lock was acquired, 0 otherwise. * If several processes call this function concurrently wanting the exclusive * lock, one will get the lock and the rest will return without getting the * lock. (If the caller must have the lock, it simply calls this function in a * loop until the function returns 1 to indicate the lock was acquired.) * Any usecnt must be decremented by calling nfsv4_relref() before * calling nfsv4_lock(). It was done this way, so nfsv4_lock() could * be called in a loop. * The isleptp argument is set to indicate if the call slept, iff not NULL * and the mp argument indicates to check for a forced dismount, iff not * NULL. */ APPLESTATIC int nfsv4_lock(struct nfsv4lock *lp, int iwantlock, int *isleptp, void *mutex, struct mount *mp) { if (isleptp) *isleptp = 0; /* * If a lock is wanted, loop around until the lock is acquired by * someone and then released. If I want the lock, try to acquire it. * For a lock to be issued, no lock must be in force and the usecnt * must be zero. */ if (iwantlock) { if (!(lp->nfslock_lock & NFSV4LOCK_LOCK) && lp->nfslock_usecnt == 0) { lp->nfslock_lock &= ~NFSV4LOCK_LOCKWANTED; lp->nfslock_lock |= NFSV4LOCK_LOCK; return (1); } lp->nfslock_lock |= NFSV4LOCK_LOCKWANTED; } while (lp->nfslock_lock & (NFSV4LOCK_LOCK | NFSV4LOCK_LOCKWANTED)) { - if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (mp != NULL && NFSCL_FORCEDISM(mp)) { lp->nfslock_lock &= ~NFSV4LOCK_LOCKWANTED; return (0); } lp->nfslock_lock |= NFSV4LOCK_WANTED; if (isleptp) *isleptp = 1; (void) nfsmsleep(&lp->nfslock_lock, mutex, PZERO - 1, "nfsv4lck", NULL); if (iwantlock && !(lp->nfslock_lock & NFSV4LOCK_LOCK) && lp->nfslock_usecnt == 0) { lp->nfslock_lock &= ~NFSV4LOCK_LOCKWANTED; lp->nfslock_lock |= NFSV4LOCK_LOCK; return (1); } } return (0); } /* * Release the lock acquired by nfsv4_lock(). * The second argument is set to 1 to indicate the nfslock_usecnt should be * incremented, as well. */ APPLESTATIC void nfsv4_unlock(struct nfsv4lock *lp, int incref) { lp->nfslock_lock &= ~NFSV4LOCK_LOCK; if (incref) lp->nfslock_usecnt++; nfsv4_wanted(lp); } /* * Release a reference cnt. */ APPLESTATIC void nfsv4_relref(struct nfsv4lock *lp) { if (lp->nfslock_usecnt <= 0) panic("nfsv4root ref cnt"); lp->nfslock_usecnt--; if (lp->nfslock_usecnt == 0) nfsv4_wanted(lp); } /* * Get a reference cnt. * This function will wait for any exclusive lock to be released, but will * not wait for threads that want the exclusive lock. If priority needs * to be given to threads that need the exclusive lock, a call to nfsv4_lock() * with the 2nd argument == 0 should be done before calling nfsv4_getref(). - * If the mp argument is not NULL, check for MNTK_UNMOUNTF being set and + * If the mp argument is not NULL, check for NFSCL_FORCEDISM() being set and * return without getting a refcnt for that case. */ APPLESTATIC void nfsv4_getref(struct nfsv4lock *lp, int *isleptp, void *mutex, struct mount *mp) { if (isleptp) *isleptp = 0; /* * Wait for a lock held. */ while (lp->nfslock_lock & NFSV4LOCK_LOCK) { - if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) + if (mp != NULL && NFSCL_FORCEDISM(mp)) return; lp->nfslock_lock |= NFSV4LOCK_WANTED; if (isleptp) *isleptp = 1; (void) nfsmsleep(&lp->nfslock_lock, mutex, PZERO - 1, "nfsv4gr", NULL); } - if (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) + if (mp != NULL && NFSCL_FORCEDISM(mp)) return; lp->nfslock_usecnt++; } /* * Get a reference as above, but return failure instead of sleeping if * an exclusive lock is held. */ APPLESTATIC int nfsv4_getref_nonblock(struct nfsv4lock *lp) { if ((lp->nfslock_lock & NFSV4LOCK_LOCK) != 0) return (0); lp->nfslock_usecnt++; return (1); } /* * Test for a lock. Return 1 if locked, 0 otherwise. */ APPLESTATIC int nfsv4_testlock(struct nfsv4lock *lp) { if ((lp->nfslock_lock & NFSV4LOCK_LOCK) == 0 && lp->nfslock_usecnt == 0) return (0); return (1); } /* * Wake up anyone sleeping, waiting for this lock. */ static void nfsv4_wanted(struct nfsv4lock *lp) { if (lp->nfslock_lock & NFSV4LOCK_WANTED) { lp->nfslock_lock &= ~NFSV4LOCK_WANTED; wakeup((caddr_t)&lp->nfslock_lock); } } /* * Copy a string from an mbuf list into a character array. * Return EBADRPC if there is an mbuf error, * 0 otherwise. */ APPLESTATIC int nfsrv_mtostr(struct nfsrv_descript *nd, char *str, int siz) { char *cp; int xfer, len; mbuf_t mp; int rem, error = 0; mp = nd->nd_md; cp = nd->nd_dpos; len = NFSMTOD(mp, caddr_t) + mbuf_len(mp) - cp; rem = NFSM_RNDUP(siz) - siz; while (siz > 0) { if (len > siz) xfer = siz; else xfer = len; NFSBCOPY(cp, str, xfer); str += xfer; siz -= xfer; if (siz > 0) { mp = mbuf_next(mp); if (mp == NULL) { error = EBADRPC; goto out; } cp = NFSMTOD(mp, caddr_t); len = mbuf_len(mp); } else { cp += xfer; len -= xfer; } } *str = '\0'; nd->nd_dpos = cp; nd->nd_md = mp; if (rem > 0) { if (len < rem) error = nfsm_advance(nd, rem, len); else nd->nd_dpos += rem; } out: NFSEXITCODE2(error, nd); return (error); } /* * Fill in the attributes as marked by the bitmap (V4). */ APPLESTATIC int nfsv4_fillattr(struct nfsrv_descript *nd, struct mount *mp, vnode_t vp, NFSACL_T *saclp, struct vattr *vap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp, struct ucred *cred, NFSPROC_T *p, int isdgram, int reterr, int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno) { int bitpos, retnum = 0; u_int32_t *tl; int siz, prefixnum, error; u_char *cp, namestr[NFSV4_SMALLSTR]; nfsattrbit_t attrbits, retbits; nfsattrbit_t *retbitp = &retbits; u_int32_t freenum, *retnump; u_int64_t uquad; struct statfs *fs; struct nfsfsinfo fsinf; struct timespec temptime; NFSACL_T *aclp, *naclp = NULL; #ifdef QUOTA struct dqblk dqb; uid_t savuid; #endif /* * First, set the bits that can be filled and get fsinfo. */ NFSSET_ATTRBIT(retbitp, attrbitp); /* * If both p and cred are NULL, it is a client side setattr call. * If both p and cred are not NULL, it is a server side reply call. * If p is not NULL and cred is NULL, it is a client side callback * reply call. */ if (p == NULL && cred == NULL) { NFSCLRNOTSETABLE_ATTRBIT(retbitp); aclp = saclp; } else { NFSCLRNOTFILLABLE_ATTRBIT(retbitp); naclp = acl_alloc(M_WAITOK); aclp = naclp; } nfsvno_getfs(&fsinf, isdgram); #ifndef APPLE /* * Get the VFS_STATFS(), since some attributes need them. */ fs = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); if (NFSISSETSTATFS_ATTRBIT(retbitp)) { error = VFS_STATFS(mp, fs); if (error != 0) { if (reterr) { nd->nd_repstat = NFSERR_ACCES; free(fs, M_STATFS); return (0); } NFSCLRSTATFS_ATTRBIT(retbitp); } } #endif /* * And the NFSv4 ACL... */ if (NFSISSET_ATTRBIT(retbitp, NFSATTRBIT_ACLSUPPORT) && (nfsrv_useacl == 0 || ((cred != NULL || p != NULL) && supports_nfsv4acls == 0))) { NFSCLRBIT_ATTRBIT(retbitp, NFSATTRBIT_ACLSUPPORT); } if (NFSISSET_ATTRBIT(retbitp, NFSATTRBIT_ACL)) { if (nfsrv_useacl == 0 || ((cred != NULL || p != NULL) && supports_nfsv4acls == 0)) { NFSCLRBIT_ATTRBIT(retbitp, NFSATTRBIT_ACL); } else if (naclp != NULL) { if (NFSVOPLOCK(vp, LK_SHARED) == 0) { error = VOP_ACCESSX(vp, VREAD_ACL, cred, p); if (error == 0) error = VOP_GETACL(vp, ACL_TYPE_NFS4, naclp, cred, p); NFSVOPUNLOCK(vp, 0); } else error = NFSERR_PERM; if (error != 0) { if (reterr) { nd->nd_repstat = NFSERR_ACCES; free(fs, M_STATFS); return (0); } NFSCLRBIT_ATTRBIT(retbitp, NFSATTRBIT_ACL); } } } /* * Put out the attribute bitmap for the ones being filled in * and get the field for the number of attributes returned. */ prefixnum = nfsrv_putattrbit(nd, retbitp); NFSM_BUILD(retnump, u_int32_t *, NFSX_UNSIGNED); prefixnum += NFSX_UNSIGNED; /* * Now, loop around filling in the attributes for each bit set. */ for (bitpos = 0; bitpos < NFSATTRBIT_MAX; bitpos++) { if (NFSISSET_ATTRBIT(retbitp, bitpos)) { switch (bitpos) { case NFSATTRBIT_SUPPORTEDATTRS: NFSSETSUPP_ATTRBIT(&attrbits); if (nfsrv_useacl == 0 || ((cred != NULL || p != NULL) && supports_nfsv4acls == 0)) { NFSCLRBIT_ATTRBIT(&attrbits,NFSATTRBIT_ACLSUPPORT); NFSCLRBIT_ATTRBIT(&attrbits,NFSATTRBIT_ACL); } retnum += nfsrv_putattrbit(nd, &attrbits); break; case NFSATTRBIT_TYPE: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = vtonfsv34_type(vap->va_type); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_FHEXPIRETYPE: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4FHTYPE_PERSISTENT); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_CHANGE: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); txdr_hyper(vap->va_filerev, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_SIZE: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); txdr_hyper(vap->va_size, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_LINKSUPPORT: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (fsinf.fs_properties & NFSV3FSINFO_LINK) *tl = newnfs_true; else *tl = newnfs_false; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_SYMLINKSUPPORT: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (fsinf.fs_properties & NFSV3FSINFO_SYMLINK) *tl = newnfs_true; else *tl = newnfs_false; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_NAMEDATTR: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_FSID: NFSM_BUILD(tl, u_int32_t *, NFSX_V4FSID); *tl++ = 0; *tl++ = txdr_unsigned(mp->mnt_stat.f_fsid.val[0]); *tl++ = 0; *tl = txdr_unsigned(mp->mnt_stat.f_fsid.val[1]); retnum += NFSX_V4FSID; break; case NFSATTRBIT_UNIQUEHANDLES: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_true; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_LEASETIME: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(nfsrv_lease); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_RDATTRERROR: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(rderror); retnum += NFSX_UNSIGNED; break; /* * Recommended Attributes. (Only the supported ones.) */ case NFSATTRBIT_ACL: retnum += nfsrv_buildacl(nd, aclp, vnode_vtype(vp), p); break; case NFSATTRBIT_ACLSUPPORT: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4ACE_SUPTYPES); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_CANSETTIME: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (fsinf.fs_properties & NFSV3FSINFO_CANSETTIME) *tl = newnfs_true; else *tl = newnfs_false; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_CASEINSENSITIVE: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_CASEPRESERVING: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_true; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_CHOWNRESTRICTED: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_true; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_FILEHANDLE: retnum += nfsm_fhtom(nd, (u_int8_t *)fhp, 0, 0); break; case NFSATTRBIT_FILEID: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = vap->va_fileid; txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_FILESAVAIL: /* * Check quota and use min(quota, f_ffree). */ freenum = fs->f_ffree; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(mp, QCMD(Q_GETQUOTA,USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = min(dqb.dqb_isoftlimit-dqb.dqb_curinodes, freenum); p->p_cred->p_ruid = savuid; #endif /* QUOTA */ NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); *tl++ = 0; *tl = txdr_unsigned(freenum); retnum += NFSX_HYPER; break; case NFSATTRBIT_FILESFREE: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); *tl++ = 0; *tl = txdr_unsigned(fs->f_ffree); retnum += NFSX_HYPER; break; case NFSATTRBIT_FILESTOTAL: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); *tl++ = 0; *tl = txdr_unsigned(fs->f_files); retnum += NFSX_HYPER; break; case NFSATTRBIT_FSLOCATIONS: NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; *tl = 0; retnum += 2 * NFSX_UNSIGNED; break; case NFSATTRBIT_HOMOGENEOUS: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (fsinf.fs_properties & NFSV3FSINFO_HOMOGENEOUS) *tl = newnfs_true; else *tl = newnfs_false; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_MAXFILESIZE: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = NFSRV_MAXFILESIZE; txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_MAXLINK: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(LINK_MAX); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_MAXNAME: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFS_MAXNAMLEN); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_MAXREAD: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); *tl++ = 0; *tl = txdr_unsigned(fsinf.fs_rtmax); retnum += NFSX_HYPER; break; case NFSATTRBIT_MAXWRITE: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); *tl++ = 0; *tl = txdr_unsigned(fsinf.fs_wtmax); retnum += NFSX_HYPER; break; case NFSATTRBIT_MODE: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = vtonfsv34_mode(vap->va_mode); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_NOTRUNC: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_true; retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_NUMLINKS: NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(vap->va_nlink); retnum += NFSX_UNSIGNED; break; case NFSATTRBIT_OWNER: cp = namestr; nfsv4_uidtostr(vap->va_uid, &cp, &siz, p); retnum += nfsm_strtom(nd, cp, siz); if (cp != namestr) free(cp, M_NFSSTRING); break; case NFSATTRBIT_OWNERGROUP: cp = namestr; nfsv4_gidtostr(vap->va_gid, &cp, &siz, p); retnum += nfsm_strtom(nd, cp, siz); if (cp != namestr) free(cp, M_NFSSTRING); break; case NFSATTRBIT_QUOTAHARD: if (priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0)) freenum = fs->f_bfree; else freenum = fs->f_bavail; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(mp, QCMD(Q_GETQUOTA,USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = min(dqb.dqb_bhardlimit, freenum); p->p_cred->p_ruid = savuid; #endif /* QUOTA */ NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = (u_int64_t)freenum; NFSQUOTABLKTOBYTE(uquad, fs->f_bsize); txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_QUOTASOFT: if (priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0)) freenum = fs->f_bfree; else freenum = fs->f_bavail; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(mp, QCMD(Q_GETQUOTA,USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = min(dqb.dqb_bsoftlimit, freenum); p->p_cred->p_ruid = savuid; #endif /* QUOTA */ NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = (u_int64_t)freenum; NFSQUOTABLKTOBYTE(uquad, fs->f_bsize); txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_QUOTAUSED: freenum = 0; #ifdef QUOTA /* * ufs_quotactl() insists that the uid argument * equal p_ruid for non-root quota access, so * we'll just make sure that's the case. */ savuid = p->p_cred->p_ruid; p->p_cred->p_ruid = cred->cr_uid; if (!VFS_QUOTACTL(mp, QCMD(Q_GETQUOTA,USRQUOTA), cred->cr_uid, (caddr_t)&dqb)) freenum = dqb.dqb_curblocks; p->p_cred->p_ruid = savuid; #endif /* QUOTA */ NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = (u_int64_t)freenum; NFSQUOTABLKTOBYTE(uquad, fs->f_bsize); txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_RAWDEV: NFSM_BUILD(tl, u_int32_t *, NFSX_V4SPECDATA); *tl++ = txdr_unsigned(NFSMAJOR(vap->va_rdev)); *tl = txdr_unsigned(NFSMINOR(vap->va_rdev)); retnum += NFSX_V4SPECDATA; break; case NFSATTRBIT_SPACEAVAIL: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0)) uquad = (u_int64_t)fs->f_bfree; else uquad = (u_int64_t)fs->f_bavail; uquad *= fs->f_bsize; txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_SPACEFREE: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = (u_int64_t)fs->f_bfree; uquad *= fs->f_bsize; txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_SPACETOTAL: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); uquad = (u_int64_t)fs->f_blocks; uquad *= fs->f_bsize; txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_SPACEUSED: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); txdr_hyper(vap->va_bytes, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_TIMEACCESS: NFSM_BUILD(tl, u_int32_t *, NFSX_V4TIME); txdr_nfsv4time(&vap->va_atime, tl); retnum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEACCESSSET: if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { NFSM_BUILD(tl, u_int32_t *, NFSX_V4SETTIME); *tl++ = txdr_unsigned(NFSV4SATTRTIME_TOCLIENT); txdr_nfsv4time(&vap->va_atime, tl); retnum += NFSX_V4SETTIME; } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4SATTRTIME_TOSERVER); retnum += NFSX_UNSIGNED; } break; case NFSATTRBIT_TIMEDELTA: NFSM_BUILD(tl, u_int32_t *, NFSX_V4TIME); temptime.tv_sec = 0; temptime.tv_nsec = 1000000000 / hz; txdr_nfsv4time(&temptime, tl); retnum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMETADATA: NFSM_BUILD(tl, u_int32_t *, NFSX_V4TIME); txdr_nfsv4time(&vap->va_ctime, tl); retnum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFY: NFSM_BUILD(tl, u_int32_t *, NFSX_V4TIME); txdr_nfsv4time(&vap->va_mtime, tl); retnum += NFSX_V4TIME; break; case NFSATTRBIT_TIMEMODIFYSET: if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { NFSM_BUILD(tl, u_int32_t *, NFSX_V4SETTIME); *tl++ = txdr_unsigned(NFSV4SATTRTIME_TOCLIENT); txdr_nfsv4time(&vap->va_mtime, tl); retnum += NFSX_V4SETTIME; } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4SATTRTIME_TOSERVER); retnum += NFSX_UNSIGNED; } break; case NFSATTRBIT_MOUNTEDONFILEID: NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER); if (at_root != 0) uquad = mounted_on_fileno; else uquad = vap->va_fileid; txdr_hyper(uquad, tl); retnum += NFSX_HYPER; break; case NFSATTRBIT_SUPPATTREXCLCREAT: NFSSETSUPP_ATTRBIT(&attrbits); NFSCLRNOTSETABLE_ATTRBIT(&attrbits); NFSCLRBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET); retnum += nfsrv_putattrbit(nd, &attrbits); break; default: printf("EEK! Bad V4 attribute bitpos=%d\n", bitpos); } } } if (naclp != NULL) acl_free(naclp); free(fs, M_STATFS); *retnump = txdr_unsigned(retnum); return (retnum + prefixnum); } /* * Put the attribute bits onto an mbuf list. * Return the number of bytes of output generated. */ APPLESTATIC int nfsrv_putattrbit(struct nfsrv_descript *nd, nfsattrbit_t *attrbitp) { u_int32_t *tl; int cnt, i, bytesize; for (cnt = NFSATTRBIT_MAXWORDS; cnt > 0; cnt--) if (attrbitp->bits[cnt - 1]) break; bytesize = (cnt + 1) * NFSX_UNSIGNED; NFSM_BUILD(tl, u_int32_t *, bytesize); *tl++ = txdr_unsigned(cnt); for (i = 0; i < cnt; i++) *tl++ = txdr_unsigned(attrbitp->bits[i]); return (bytesize); } /* * Convert a uid to a string. * If the lookup fails, just output the digits. * uid - the user id * cpp - points to a buffer of size NFSV4_SMALLSTR * (malloc a larger one, as required) * retlenp - pointer to length to be returned */ APPLESTATIC void nfsv4_uidtostr(uid_t uid, u_char **cpp, int *retlenp, NFSPROC_T *p) { int i; struct nfsusrgrp *usrp; u_char *cp = *cpp; uid_t tmp; int cnt, hasampersand, len = NFSV4_SMALLSTR, ret; struct nfsrv_lughash *hp; cnt = 0; tryagain: if (nfsrv_dnsnamelen > 0 && !nfs_enable_uidtostring) { /* * Always map nfsrv_defaultuid to "nobody". */ if (uid == nfsrv_defaultuid) { i = nfsrv_dnsnamelen + 7; if (i > len) { if (len > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); cp = malloc(i, M_NFSSTRING, M_WAITOK); *cpp = cp; len = i; goto tryagain; } *retlenp = i; NFSBCOPY("nobody@", cp, 7); cp += 7; NFSBCOPY(nfsrv_dnsname, cp, nfsrv_dnsnamelen); return; } hasampersand = 0; hp = NFSUSERHASH(uid); mtx_lock(&hp->mtx); TAILQ_FOREACH(usrp, &hp->lughead, lug_numhash) { if (usrp->lug_uid == uid) { if (usrp->lug_expiry < NFSD_MONOSEC) break; /* * If the name doesn't already have an '@' * in it, append @domainname to it. */ for (i = 0; i < usrp->lug_namelen; i++) { if (usrp->lug_name[i] == '@') { hasampersand = 1; break; } } if (hasampersand) i = usrp->lug_namelen; else i = usrp->lug_namelen + nfsrv_dnsnamelen + 1; if (i > len) { mtx_unlock(&hp->mtx); if (len > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); cp = malloc(i, M_NFSSTRING, M_WAITOK); *cpp = cp; len = i; goto tryagain; } *retlenp = i; NFSBCOPY(usrp->lug_name, cp, usrp->lug_namelen); if (!hasampersand) { cp += usrp->lug_namelen; *cp++ = '@'; NFSBCOPY(nfsrv_dnsname, cp, nfsrv_dnsnamelen); } TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); TAILQ_INSERT_TAIL(&hp->lughead, usrp, lug_numhash); mtx_unlock(&hp->mtx); return; } } mtx_unlock(&hp->mtx); cnt++; ret = nfsrv_getuser(RPCNFSUSERD_GETUID, uid, (gid_t)0, NULL, p); if (ret == 0 && cnt < 2) goto tryagain; } /* * No match, just return a string of digits. */ tmp = uid; i = 0; while (tmp || i == 0) { tmp /= 10; i++; } len = (i > len) ? len : i; *retlenp = len; cp += (len - 1); tmp = uid; for (i = 0; i < len; i++) { *cp-- = '0' + (tmp % 10); tmp /= 10; } return; } /* * Get a credential for the uid with the server's group list. * If none is found, just return the credential passed in after * logging a warning message. */ struct ucred * nfsrv_getgrpscred(struct ucred *oldcred) { struct nfsusrgrp *usrp; struct ucred *newcred; int cnt, ret; uid_t uid; struct nfsrv_lughash *hp; cnt = 0; uid = oldcred->cr_uid; tryagain: if (nfsrv_dnsnamelen > 0) { hp = NFSUSERHASH(uid); mtx_lock(&hp->mtx); TAILQ_FOREACH(usrp, &hp->lughead, lug_numhash) { if (usrp->lug_uid == uid) { if (usrp->lug_expiry < NFSD_MONOSEC) break; if (usrp->lug_cred != NULL) { newcred = crhold(usrp->lug_cred); crfree(oldcred); } else newcred = oldcred; TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); TAILQ_INSERT_TAIL(&hp->lughead, usrp, lug_numhash); mtx_unlock(&hp->mtx); return (newcred); } } mtx_unlock(&hp->mtx); cnt++; ret = nfsrv_getuser(RPCNFSUSERD_GETUID, uid, (gid_t)0, NULL, curthread); if (ret == 0 && cnt < 2) goto tryagain; } return (oldcred); } /* * Convert a string to a uid. * If no conversion is possible return NFSERR_BADOWNER, otherwise * return 0. * If this is called from a client side mount using AUTH_SYS and the * string is made up entirely of digits, just convert the string to * a number. */ APPLESTATIC int nfsv4_strtouid(struct nfsrv_descript *nd, u_char *str, int len, uid_t *uidp, NFSPROC_T *p) { int i; char *cp, *endstr, *str0; struct nfsusrgrp *usrp; int cnt, ret; int error = 0; uid_t tuid; struct nfsrv_lughash *hp, *hp2; if (len == 0) { error = NFSERR_BADOWNER; goto out; } /* If a string of digits and an AUTH_SYS mount, just convert it. */ str0 = str; tuid = (uid_t)strtoul(str0, &endstr, 10); if ((endstr - str0) == len) { /* A numeric string. */ if ((nd->nd_flag & ND_KERBV) == 0 && ((nd->nd_flag & ND_NFSCL) != 0 || nfsd_enable_stringtouid != 0)) *uidp = tuid; else error = NFSERR_BADOWNER; goto out; } /* * Look for an '@'. */ cp = strchr(str0, '@'); if (cp != NULL) i = (int)(cp++ - str0); else i = len; cnt = 0; tryagain: if (nfsrv_dnsnamelen > 0) { /* * If an '@' is found and the domain name matches, search for * the name with dns stripped off. * Mixed case alpahbetics will match for the domain name, but * all upper case will not. */ if (cnt == 0 && i < len && i > 0 && (len - 1 - i) == nfsrv_dnsnamelen && !nfsrv_cmpmixedcase(cp, nfsrv_dnsname, nfsrv_dnsnamelen)) { len -= (nfsrv_dnsnamelen + 1); *(cp - 1) = '\0'; } /* * Check for the special case of "nobody". */ if (len == 6 && !NFSBCMP(str, "nobody", 6)) { *uidp = nfsrv_defaultuid; error = 0; goto out; } hp = NFSUSERNAMEHASH(str, len); mtx_lock(&hp->mtx); TAILQ_FOREACH(usrp, &hp->lughead, lug_namehash) { if (usrp->lug_namelen == len && !NFSBCMP(usrp->lug_name, str, len)) { if (usrp->lug_expiry < NFSD_MONOSEC) break; hp2 = NFSUSERHASH(usrp->lug_uid); mtx_lock(&hp2->mtx); TAILQ_REMOVE(&hp2->lughead, usrp, lug_numhash); TAILQ_INSERT_TAIL(&hp2->lughead, usrp, lug_numhash); *uidp = usrp->lug_uid; mtx_unlock(&hp2->mtx); mtx_unlock(&hp->mtx); error = 0; goto out; } } mtx_unlock(&hp->mtx); cnt++; ret = nfsrv_getuser(RPCNFSUSERD_GETUSER, (uid_t)0, (gid_t)0, str, p); if (ret == 0 && cnt < 2) goto tryagain; } error = NFSERR_BADOWNER; out: NFSEXITCODE(error); return (error); } /* * Convert a gid to a string. * gid - the group id * cpp - points to a buffer of size NFSV4_SMALLSTR * (malloc a larger one, as required) * retlenp - pointer to length to be returned */ APPLESTATIC void nfsv4_gidtostr(gid_t gid, u_char **cpp, int *retlenp, NFSPROC_T *p) { int i; struct nfsusrgrp *usrp; u_char *cp = *cpp; gid_t tmp; int cnt, hasampersand, len = NFSV4_SMALLSTR, ret; struct nfsrv_lughash *hp; cnt = 0; tryagain: if (nfsrv_dnsnamelen > 0 && !nfs_enable_uidtostring) { /* * Always map nfsrv_defaultgid to "nogroup". */ if (gid == nfsrv_defaultgid) { i = nfsrv_dnsnamelen + 8; if (i > len) { if (len > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); cp = malloc(i, M_NFSSTRING, M_WAITOK); *cpp = cp; len = i; goto tryagain; } *retlenp = i; NFSBCOPY("nogroup@", cp, 8); cp += 8; NFSBCOPY(nfsrv_dnsname, cp, nfsrv_dnsnamelen); return; } hasampersand = 0; hp = NFSGROUPHASH(gid); mtx_lock(&hp->mtx); TAILQ_FOREACH(usrp, &hp->lughead, lug_numhash) { if (usrp->lug_gid == gid) { if (usrp->lug_expiry < NFSD_MONOSEC) break; /* * If the name doesn't already have an '@' * in it, append @domainname to it. */ for (i = 0; i < usrp->lug_namelen; i++) { if (usrp->lug_name[i] == '@') { hasampersand = 1; break; } } if (hasampersand) i = usrp->lug_namelen; else i = usrp->lug_namelen + nfsrv_dnsnamelen + 1; if (i > len) { mtx_unlock(&hp->mtx); if (len > NFSV4_SMALLSTR) free(cp, M_NFSSTRING); cp = malloc(i, M_NFSSTRING, M_WAITOK); *cpp = cp; len = i; goto tryagain; } *retlenp = i; NFSBCOPY(usrp->lug_name, cp, usrp->lug_namelen); if (!hasampersand) { cp += usrp->lug_namelen; *cp++ = '@'; NFSBCOPY(nfsrv_dnsname, cp, nfsrv_dnsnamelen); } TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); TAILQ_INSERT_TAIL(&hp->lughead, usrp, lug_numhash); mtx_unlock(&hp->mtx); return; } } mtx_unlock(&hp->mtx); cnt++; ret = nfsrv_getuser(RPCNFSUSERD_GETGID, (uid_t)0, gid, NULL, p); if (ret == 0 && cnt < 2) goto tryagain; } /* * No match, just return a string of digits. */ tmp = gid; i = 0; while (tmp || i == 0) { tmp /= 10; i++; } len = (i > len) ? len : i; *retlenp = len; cp += (len - 1); tmp = gid; for (i = 0; i < len; i++) { *cp-- = '0' + (tmp % 10); tmp /= 10; } return; } /* * Convert a string to a gid. * If no conversion is possible return NFSERR_BADOWNER, otherwise * return 0. * If this is called from a client side mount using AUTH_SYS and the * string is made up entirely of digits, just convert the string to * a number. */ APPLESTATIC int nfsv4_strtogid(struct nfsrv_descript *nd, u_char *str, int len, gid_t *gidp, NFSPROC_T *p) { int i; char *cp, *endstr, *str0; struct nfsusrgrp *usrp; int cnt, ret; int error = 0; gid_t tgid; struct nfsrv_lughash *hp, *hp2; if (len == 0) { error = NFSERR_BADOWNER; goto out; } /* If a string of digits and an AUTH_SYS mount, just convert it. */ str0 = str; tgid = (gid_t)strtoul(str0, &endstr, 10); if ((endstr - str0) == len) { /* A numeric string. */ if ((nd->nd_flag & ND_KERBV) == 0 && ((nd->nd_flag & ND_NFSCL) != 0 || nfsd_enable_stringtouid != 0)) *gidp = tgid; else error = NFSERR_BADOWNER; goto out; } /* * Look for an '@'. */ cp = strchr(str0, '@'); if (cp != NULL) i = (int)(cp++ - str0); else i = len; cnt = 0; tryagain: if (nfsrv_dnsnamelen > 0) { /* * If an '@' is found and the dns name matches, search for the * name with the dns stripped off. */ if (cnt == 0 && i < len && i > 0 && (len - 1 - i) == nfsrv_dnsnamelen && !nfsrv_cmpmixedcase(cp, nfsrv_dnsname, nfsrv_dnsnamelen)) { len -= (nfsrv_dnsnamelen + 1); *(cp - 1) = '\0'; } /* * Check for the special case of "nogroup". */ if (len == 7 && !NFSBCMP(str, "nogroup", 7)) { *gidp = nfsrv_defaultgid; error = 0; goto out; } hp = NFSGROUPNAMEHASH(str, len); mtx_lock(&hp->mtx); TAILQ_FOREACH(usrp, &hp->lughead, lug_namehash) { if (usrp->lug_namelen == len && !NFSBCMP(usrp->lug_name, str, len)) { if (usrp->lug_expiry < NFSD_MONOSEC) break; hp2 = NFSGROUPHASH(usrp->lug_gid); mtx_lock(&hp2->mtx); TAILQ_REMOVE(&hp2->lughead, usrp, lug_numhash); TAILQ_INSERT_TAIL(&hp2->lughead, usrp, lug_numhash); *gidp = usrp->lug_gid; mtx_unlock(&hp2->mtx); mtx_unlock(&hp->mtx); error = 0; goto out; } } mtx_unlock(&hp->mtx); cnt++; ret = nfsrv_getuser(RPCNFSUSERD_GETGROUP, (uid_t)0, (gid_t)0, str, p); if (ret == 0 && cnt < 2) goto tryagain; } error = NFSERR_BADOWNER; out: NFSEXITCODE(error); return (error); } /* * Cmp len chars, allowing mixed case in the first argument to match lower * case in the second, but not if the first argument is all upper case. * Return 0 for a match, 1 otherwise. */ static int nfsrv_cmpmixedcase(u_char *cp, u_char *cp2, int len) { int i; u_char tmp; int fndlower = 0; for (i = 0; i < len; i++) { if (*cp >= 'A' && *cp <= 'Z') { tmp = *cp++ + ('a' - 'A'); } else { tmp = *cp++; if (tmp >= 'a' && tmp <= 'z') fndlower = 1; } if (tmp != *cp2++) return (1); } if (fndlower) return (0); else return (1); } /* * Set the port for the nfsuserd. */ APPLESTATIC int nfsrv_nfsuserdport(struct sockaddr *sad, u_short port, NFSPROC_T *p) { struct nfssockreq *rp; struct sockaddr_in *ad; int error; NFSLOCKNAMEID(); if (nfsrv_nfsuserd) { NFSUNLOCKNAMEID(); error = EPERM; NFSSOCKADDRFREE(sad); goto out; } nfsrv_nfsuserd = 1; NFSUNLOCKNAMEID(); /* * Set up the socket record and connect. */ rp = &nfsrv_nfsuserdsock; rp->nr_client = NULL; rp->nr_cred = NULL; rp->nr_lock = (NFSR_RESERVEDPORT | NFSR_LOCALHOST); if (sad != NULL) { /* Use the AF_LOCAL socket address passed in. */ rp->nr_sotype = SOCK_STREAM; rp->nr_soproto = 0; rp->nr_nam = sad; } else { /* Use the port# for a UDP socket (old nfsuserd). */ rp->nr_sotype = SOCK_DGRAM; rp->nr_soproto = IPPROTO_UDP; NFSSOCKADDRALLOC(rp->nr_nam); NFSSOCKADDRSIZE(rp->nr_nam, sizeof (struct sockaddr_in)); ad = NFSSOCKADDR(rp->nr_nam, struct sockaddr_in *); ad->sin_family = AF_INET; ad->sin_addr.s_addr = htonl((u_int32_t)0x7f000001); ad->sin_port = port; } rp->nr_prog = RPCPROG_NFSUSERD; rp->nr_vers = RPCNFSUSERD_VERS; error = newnfs_connect(NULL, rp, NFSPROCCRED(p), p, 0); if (error) { NFSSOCKADDRFREE(rp->nr_nam); nfsrv_nfsuserd = 0; } out: NFSEXITCODE(error); return (error); } /* * Delete the nfsuserd port. */ APPLESTATIC void nfsrv_nfsuserddelport(void) { NFSLOCKNAMEID(); if (nfsrv_nfsuserd == 0) { NFSUNLOCKNAMEID(); return; } nfsrv_nfsuserd = 0; NFSUNLOCKNAMEID(); newnfs_disconnect(&nfsrv_nfsuserdsock); NFSSOCKADDRFREE(nfsrv_nfsuserdsock.nr_nam); } /* * Do upcalls to the nfsuserd, for cache misses of the owner/ownergroup * name<-->id cache. * Returns 0 upon success, non-zero otherwise. */ static int nfsrv_getuser(int procnum, uid_t uid, gid_t gid, char *name, NFSPROC_T *p) { u_int32_t *tl; struct nfsrv_descript *nd; int len; struct nfsrv_descript nfsd; struct ucred *cred; int error; NFSLOCKNAMEID(); if (nfsrv_nfsuserd == 0) { NFSUNLOCKNAMEID(); error = EPERM; goto out; } NFSUNLOCKNAMEID(); nd = &nfsd; cred = newnfs_getcred(); nd->nd_flag = ND_GSSINITREPLY; nfsrvd_rephead(nd); nd->nd_procnum = procnum; if (procnum == RPCNFSUSERD_GETUID || procnum == RPCNFSUSERD_GETGID) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (procnum == RPCNFSUSERD_GETUID) *tl = txdr_unsigned(uid); else *tl = txdr_unsigned(gid); } else { len = strlen(name); (void) nfsm_strtom(nd, name, len); } error = newnfs_request(nd, NULL, NULL, &nfsrv_nfsuserdsock, NULL, NULL, cred, RPCPROG_NFSUSERD, RPCNFSUSERD_VERS, NULL, 0, NULL, NULL); NFSFREECRED(cred); if (!error) { mbuf_freem(nd->nd_mrep); error = nd->nd_repstat; } out: NFSEXITCODE(error); return (error); } /* * This function is called from the nfssvc(2) system call, to update the * kernel user/group name list(s) for the V4 owner and ownergroup attributes. */ APPLESTATIC int nfssvc_idname(struct nfsd_idargs *nidp) { struct nfsusrgrp *nusrp, *usrp, *newusrp; struct nfsrv_lughash *hp_name, *hp_idnum, *thp; int i, group_locked, groupname_locked, user_locked, username_locked; int error = 0; u_char *cp; gid_t *grps; struct ucred *cr; static int onethread = 0; static time_t lasttime = 0; if (nidp->nid_namelen <= 0 || nidp->nid_namelen > MAXHOSTNAMELEN) { error = EINVAL; goto out; } if (nidp->nid_flag & NFSID_INITIALIZE) { cp = malloc(nidp->nid_namelen + 1, M_NFSSTRING, M_WAITOK); error = copyin(CAST_USER_ADDR_T(nidp->nid_name), cp, nidp->nid_namelen); if (error != 0) { free(cp, M_NFSSTRING); goto out; } if (atomic_cmpset_acq_int(&nfsrv_dnsnamelen, 0, 0) == 0) { /* * Free up all the old stuff and reinitialize hash * lists. All mutexes for both lists must be locked, * with the user/group name ones before the uid/gid * ones, to avoid a LOR. */ for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsusernamehash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsuserhash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) TAILQ_FOREACH_SAFE(usrp, &nfsuserhash[i].lughead, lug_numhash, nusrp) nfsrv_removeuser(usrp, 1); for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsuserhash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsusernamehash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsgroupnamehash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsgrouphash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) TAILQ_FOREACH_SAFE(usrp, &nfsgrouphash[i].lughead, lug_numhash, nusrp) nfsrv_removeuser(usrp, 0); for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsgrouphash[i].mtx); for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsgroupnamehash[i].mtx); free(nfsrv_dnsname, M_NFSSTRING); nfsrv_dnsname = NULL; } if (nfsuserhash == NULL) { /* Allocate the hash tables. */ nfsuserhash = malloc(sizeof(struct nfsrv_lughash) * nfsrv_lughashsize, M_NFSUSERGROUP, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_lughashsize; i++) mtx_init(&nfsuserhash[i].mtx, "nfsuidhash", NULL, MTX_DEF | MTX_DUPOK); nfsusernamehash = malloc(sizeof(struct nfsrv_lughash) * nfsrv_lughashsize, M_NFSUSERGROUP, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_lughashsize; i++) mtx_init(&nfsusernamehash[i].mtx, "nfsusrhash", NULL, MTX_DEF | MTX_DUPOK); nfsgrouphash = malloc(sizeof(struct nfsrv_lughash) * nfsrv_lughashsize, M_NFSUSERGROUP, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_lughashsize; i++) mtx_init(&nfsgrouphash[i].mtx, "nfsgidhash", NULL, MTX_DEF | MTX_DUPOK); nfsgroupnamehash = malloc(sizeof(struct nfsrv_lughash) * nfsrv_lughashsize, M_NFSUSERGROUP, M_WAITOK | M_ZERO); for (i = 0; i < nfsrv_lughashsize; i++) mtx_init(&nfsgroupnamehash[i].mtx, "nfsgrphash", NULL, MTX_DEF | MTX_DUPOK); } /* (Re)initialize the list heads. */ for (i = 0; i < nfsrv_lughashsize; i++) TAILQ_INIT(&nfsuserhash[i].lughead); for (i = 0; i < nfsrv_lughashsize; i++) TAILQ_INIT(&nfsusernamehash[i].lughead); for (i = 0; i < nfsrv_lughashsize; i++) TAILQ_INIT(&nfsgrouphash[i].lughead); for (i = 0; i < nfsrv_lughashsize; i++) TAILQ_INIT(&nfsgroupnamehash[i].lughead); /* * Put name in "DNS" string. */ nfsrv_dnsname = cp; nfsrv_defaultuid = nidp->nid_uid; nfsrv_defaultgid = nidp->nid_gid; nfsrv_usercnt = 0; nfsrv_usermax = nidp->nid_usermax; atomic_store_rel_int(&nfsrv_dnsnamelen, nidp->nid_namelen); goto out; } /* * malloc the new one now, so any potential sleep occurs before * manipulation of the lists. */ newusrp = malloc(sizeof(struct nfsusrgrp) + nidp->nid_namelen, M_NFSUSERGROUP, M_WAITOK | M_ZERO); error = copyin(CAST_USER_ADDR_T(nidp->nid_name), newusrp->lug_name, nidp->nid_namelen); if (error == 0 && nidp->nid_ngroup > 0 && (nidp->nid_flag & NFSID_ADDUID) != 0) { grps = malloc(sizeof(gid_t) * nidp->nid_ngroup, M_TEMP, M_WAITOK); error = copyin(CAST_USER_ADDR_T(nidp->nid_grps), grps, sizeof(gid_t) * nidp->nid_ngroup); if (error == 0) { /* * Create a credential just like svc_getcred(), * but using the group list provided. */ cr = crget(); cr->cr_uid = cr->cr_ruid = cr->cr_svuid = nidp->nid_uid; crsetgroups(cr, nidp->nid_ngroup, grps); cr->cr_rgid = cr->cr_svgid = cr->cr_groups[0]; cr->cr_prison = &prison0; prison_hold(cr->cr_prison); #ifdef MAC mac_cred_associate_nfsd(cr); #endif newusrp->lug_cred = cr; } free(grps, M_TEMP); } if (error) { free(newusrp, M_NFSUSERGROUP); goto out; } newusrp->lug_namelen = nidp->nid_namelen; /* * The lock order is username[0]->[nfsrv_lughashsize - 1] followed * by uid[0]->[nfsrv_lughashsize - 1], with the same for group. * The flags user_locked, username_locked, group_locked and * groupname_locked are set to indicate all of those hash lists are * locked. hp_name != NULL and hp_idnum != NULL indicates that * the respective one mutex is locked. */ user_locked = username_locked = group_locked = groupname_locked = 0; hp_name = hp_idnum = NULL; /* * Delete old entries, as required. */ if (nidp->nid_flag & (NFSID_DELUID | NFSID_ADDUID)) { /* Must lock all username hash lists first, to avoid a LOR. */ for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsusernamehash[i].mtx); username_locked = 1; hp_idnum = NFSUSERHASH(nidp->nid_uid); mtx_lock(&hp_idnum->mtx); TAILQ_FOREACH_SAFE(usrp, &hp_idnum->lughead, lug_numhash, nusrp) { if (usrp->lug_uid == nidp->nid_uid) nfsrv_removeuser(usrp, 1); } } else if (nidp->nid_flag & (NFSID_DELUSERNAME | NFSID_ADDUSERNAME)) { hp_name = NFSUSERNAMEHASH(newusrp->lug_name, newusrp->lug_namelen); mtx_lock(&hp_name->mtx); TAILQ_FOREACH_SAFE(usrp, &hp_name->lughead, lug_namehash, nusrp) { if (usrp->lug_namelen == newusrp->lug_namelen && !NFSBCMP(usrp->lug_name, newusrp->lug_name, usrp->lug_namelen)) { thp = NFSUSERHASH(usrp->lug_uid); mtx_lock(&thp->mtx); nfsrv_removeuser(usrp, 1); mtx_unlock(&thp->mtx); } } hp_idnum = NFSUSERHASH(nidp->nid_uid); mtx_lock(&hp_idnum->mtx); } else if (nidp->nid_flag & (NFSID_DELGID | NFSID_ADDGID)) { /* Must lock all groupname hash lists first, to avoid a LOR. */ for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsgroupnamehash[i].mtx); groupname_locked = 1; hp_idnum = NFSGROUPHASH(nidp->nid_gid); mtx_lock(&hp_idnum->mtx); TAILQ_FOREACH_SAFE(usrp, &hp_idnum->lughead, lug_numhash, nusrp) { if (usrp->lug_gid == nidp->nid_gid) nfsrv_removeuser(usrp, 0); } } else if (nidp->nid_flag & (NFSID_DELGROUPNAME | NFSID_ADDGROUPNAME)) { hp_name = NFSGROUPNAMEHASH(newusrp->lug_name, newusrp->lug_namelen); mtx_lock(&hp_name->mtx); TAILQ_FOREACH_SAFE(usrp, &hp_name->lughead, lug_namehash, nusrp) { if (usrp->lug_namelen == newusrp->lug_namelen && !NFSBCMP(usrp->lug_name, newusrp->lug_name, usrp->lug_namelen)) { thp = NFSGROUPHASH(usrp->lug_gid); mtx_lock(&thp->mtx); nfsrv_removeuser(usrp, 0); mtx_unlock(&thp->mtx); } } hp_idnum = NFSGROUPHASH(nidp->nid_gid); mtx_lock(&hp_idnum->mtx); } /* * Now, we can add the new one. */ if (nidp->nid_usertimeout) newusrp->lug_expiry = NFSD_MONOSEC + nidp->nid_usertimeout; else newusrp->lug_expiry = NFSD_MONOSEC + 5; if (nidp->nid_flag & (NFSID_ADDUID | NFSID_ADDUSERNAME)) { newusrp->lug_uid = nidp->nid_uid; thp = NFSUSERHASH(newusrp->lug_uid); mtx_assert(&thp->mtx, MA_OWNED); TAILQ_INSERT_TAIL(&thp->lughead, newusrp, lug_numhash); thp = NFSUSERNAMEHASH(newusrp->lug_name, newusrp->lug_namelen); mtx_assert(&thp->mtx, MA_OWNED); TAILQ_INSERT_TAIL(&thp->lughead, newusrp, lug_namehash); atomic_add_int(&nfsrv_usercnt, 1); } else if (nidp->nid_flag & (NFSID_ADDGID | NFSID_ADDGROUPNAME)) { newusrp->lug_gid = nidp->nid_gid; thp = NFSGROUPHASH(newusrp->lug_gid); mtx_assert(&thp->mtx, MA_OWNED); TAILQ_INSERT_TAIL(&thp->lughead, newusrp, lug_numhash); thp = NFSGROUPNAMEHASH(newusrp->lug_name, newusrp->lug_namelen); mtx_assert(&thp->mtx, MA_OWNED); TAILQ_INSERT_TAIL(&thp->lughead, newusrp, lug_namehash); atomic_add_int(&nfsrv_usercnt, 1); } else { if (newusrp->lug_cred != NULL) crfree(newusrp->lug_cred); free(newusrp, M_NFSUSERGROUP); } /* * Once per second, allow one thread to trim the cache. */ if (lasttime < NFSD_MONOSEC && atomic_cmpset_acq_int(&onethread, 0, 1) != 0) { /* * First, unlock the single mutexes, so that all entries * can be locked and any LOR is avoided. */ if (hp_name != NULL) { mtx_unlock(&hp_name->mtx); hp_name = NULL; } if (hp_idnum != NULL) { mtx_unlock(&hp_idnum->mtx); hp_idnum = NULL; } if ((nidp->nid_flag & (NFSID_DELUID | NFSID_ADDUID | NFSID_DELUSERNAME | NFSID_ADDUSERNAME)) != 0) { if (username_locked == 0) { for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsusernamehash[i].mtx); username_locked = 1; } KASSERT(user_locked == 0, ("nfssvc_idname: user_locked")); for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsuserhash[i].mtx); user_locked = 1; for (i = 0; i < nfsrv_lughashsize; i++) { TAILQ_FOREACH_SAFE(usrp, &nfsuserhash[i].lughead, lug_numhash, nusrp) if (usrp->lug_expiry < NFSD_MONOSEC) nfsrv_removeuser(usrp, 1); } for (i = 0; i < nfsrv_lughashsize; i++) { /* * Trim the cache using an approximate LRU * algorithm. This code deletes the least * recently used entry on each hash list. */ if (nfsrv_usercnt <= nfsrv_usermax) break; usrp = TAILQ_FIRST(&nfsuserhash[i].lughead); if (usrp != NULL) nfsrv_removeuser(usrp, 1); } } else { if (groupname_locked == 0) { for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsgroupnamehash[i].mtx); groupname_locked = 1; } KASSERT(group_locked == 0, ("nfssvc_idname: group_locked")); for (i = 0; i < nfsrv_lughashsize; i++) mtx_lock(&nfsgrouphash[i].mtx); group_locked = 1; for (i = 0; i < nfsrv_lughashsize; i++) { TAILQ_FOREACH_SAFE(usrp, &nfsgrouphash[i].lughead, lug_numhash, nusrp) if (usrp->lug_expiry < NFSD_MONOSEC) nfsrv_removeuser(usrp, 0); } for (i = 0; i < nfsrv_lughashsize; i++) { /* * Trim the cache using an approximate LRU * algorithm. This code deletes the least * recently user entry on each hash list. */ if (nfsrv_usercnt <= nfsrv_usermax) break; usrp = TAILQ_FIRST(&nfsgrouphash[i].lughead); if (usrp != NULL) nfsrv_removeuser(usrp, 0); } } lasttime = NFSD_MONOSEC; atomic_store_rel_int(&onethread, 0); } /* Now, unlock all locked mutexes. */ if (hp_idnum != NULL) mtx_unlock(&hp_idnum->mtx); if (hp_name != NULL) mtx_unlock(&hp_name->mtx); if (user_locked != 0) for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsuserhash[i].mtx); if (username_locked != 0) for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsusernamehash[i].mtx); if (group_locked != 0) for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsgrouphash[i].mtx); if (groupname_locked != 0) for (i = 0; i < nfsrv_lughashsize; i++) mtx_unlock(&nfsgroupnamehash[i].mtx); out: NFSEXITCODE(error); return (error); } /* * Remove a user/group name element. */ static void nfsrv_removeuser(struct nfsusrgrp *usrp, int isuser) { struct nfsrv_lughash *hp; if (isuser != 0) { hp = NFSUSERHASH(usrp->lug_uid); mtx_assert(&hp->mtx, MA_OWNED); TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); hp = NFSUSERNAMEHASH(usrp->lug_name, usrp->lug_namelen); mtx_assert(&hp->mtx, MA_OWNED); TAILQ_REMOVE(&hp->lughead, usrp, lug_namehash); } else { hp = NFSGROUPHASH(usrp->lug_gid); mtx_assert(&hp->mtx, MA_OWNED); TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); hp = NFSGROUPNAMEHASH(usrp->lug_name, usrp->lug_namelen); mtx_assert(&hp->mtx, MA_OWNED); TAILQ_REMOVE(&hp->lughead, usrp, lug_namehash); } atomic_add_int(&nfsrv_usercnt, -1); if (usrp->lug_cred != NULL) crfree(usrp->lug_cred); free(usrp, M_NFSUSERGROUP); } /* * Free up all the allocations related to the name<-->id cache. * This function should only be called when the nfsuserd daemon isn't * running, since it doesn't do any locking. * This function is meant to be used when the nfscommon module is unloaded. */ APPLESTATIC void nfsrv_cleanusergroup(void) { struct nfsrv_lughash *hp, *hp2; struct nfsusrgrp *nusrp, *usrp; int i; if (nfsuserhash == NULL) return; for (i = 0; i < nfsrv_lughashsize; i++) { hp = &nfsuserhash[i]; TAILQ_FOREACH_SAFE(usrp, &hp->lughead, lug_numhash, nusrp) { TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); hp2 = NFSUSERNAMEHASH(usrp->lug_name, usrp->lug_namelen); TAILQ_REMOVE(&hp2->lughead, usrp, lug_namehash); if (usrp->lug_cred != NULL) crfree(usrp->lug_cred); free(usrp, M_NFSUSERGROUP); } hp = &nfsgrouphash[i]; TAILQ_FOREACH_SAFE(usrp, &hp->lughead, lug_numhash, nusrp) { TAILQ_REMOVE(&hp->lughead, usrp, lug_numhash); hp2 = NFSGROUPNAMEHASH(usrp->lug_name, usrp->lug_namelen); TAILQ_REMOVE(&hp2->lughead, usrp, lug_namehash); if (usrp->lug_cred != NULL) crfree(usrp->lug_cred); free(usrp, M_NFSUSERGROUP); } mtx_destroy(&nfsuserhash[i].mtx); mtx_destroy(&nfsusernamehash[i].mtx); mtx_destroy(&nfsgroupnamehash[i].mtx); mtx_destroy(&nfsgrouphash[i].mtx); } free(nfsuserhash, M_NFSUSERGROUP); free(nfsusernamehash, M_NFSUSERGROUP); free(nfsgrouphash, M_NFSUSERGROUP); free(nfsgroupnamehash, M_NFSUSERGROUP); free(nfsrv_dnsname, M_NFSSTRING); } /* * This function scans a byte string and checks for UTF-8 compliance. * It returns 0 if it conforms and NFSERR_INVAL if not. */ APPLESTATIC int nfsrv_checkutf8(u_int8_t *cp, int len) { u_int32_t val = 0x0; int cnt = 0, gotd = 0, shift = 0; u_int8_t byte; static int utf8_shift[5] = { 7, 11, 16, 21, 26 }; int error = 0; /* * Here are what the variables are used for: * val - the calculated value of a multibyte char, used to check * that it was coded with the correct range * cnt - the number of 10xxxxxx bytes to follow * gotd - set for a char of Dxxx, so D800<->DFFF can be checked for * shift - lower order bits of range (ie. "val >> shift" should * not be 0, in other words, dividing by the lower bound * of the range should get a non-zero value) * byte - used to calculate cnt */ while (len > 0) { if (cnt > 0) { /* This handles the 10xxxxxx bytes */ if ((*cp & 0xc0) != 0x80 || (gotd && (*cp & 0x20))) { error = NFSERR_INVAL; goto out; } gotd = 0; val <<= 6; val |= (*cp & 0x3f); cnt--; if (cnt == 0 && (val >> shift) == 0x0) { error = NFSERR_INVAL; goto out; } } else if (*cp & 0x80) { /* first byte of multi byte char */ byte = *cp; while ((byte & 0x40) && cnt < 6) { cnt++; byte <<= 1; } if (cnt == 0 || cnt == 6) { error = NFSERR_INVAL; goto out; } val = (*cp & (0x3f >> cnt)); shift = utf8_shift[cnt - 1]; if (cnt == 2 && val == 0xd) /* Check for the 0xd800-0xdfff case */ gotd = 1; } cp++; len--; } if (cnt > 0) error = NFSERR_INVAL; out: NFSEXITCODE(error); return (error); } /* * Parse the xdr for an NFSv4 FsLocations attribute. Return two malloc'd * strings, one with the root path in it and the other with the list of * locations. The list is in the same format as is found in nfr_refs. * It is a "," separated list of entries, where each of them is of the * form :. For example * "nfsv4-test:/sub2,nfsv4-test2:/user/mnt,nfsv4-test2:/user/mnt2" * The nilp argument is set to 1 for the special case of a null fs_root * and an empty server list. * It returns NFSERR_BADXDR, if the xdr can't be parsed and returns the * number of xdr bytes parsed in sump. */ static int nfsrv_getrefstr(struct nfsrv_descript *nd, u_char **fsrootp, u_char **srvp, int *sump, int *nilp) { u_int32_t *tl; u_char *cp = NULL, *cp2 = NULL, *cp3, *str; int i, j, len, stringlen, cnt, slen, siz, xdrsum, error = 0, nsrv; struct list { SLIST_ENTRY(list) next; int len; u_char host[1]; } *lsp, *nlsp; SLIST_HEAD(, list) head; *fsrootp = NULL; *srvp = NULL; *nilp = 0; /* * Get the fs_root path and check for the special case of null path * and 0 length server list. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len < 0 || len > 10240) { error = NFSERR_BADXDR; goto nfsmout; } if (len == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl != 0) { error = NFSERR_BADXDR; goto nfsmout; } *nilp = 1; *sump = 2 * NFSX_UNSIGNED; error = 0; goto nfsmout; } cp = malloc(len + 1, M_NFSSTRING, M_WAITOK); error = nfsrv_mtostr(nd, cp, len); if (!error) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); cnt = fxdr_unsigned(int, *tl); if (cnt <= 0) error = NFSERR_BADXDR; } if (error) goto nfsmout; /* * Now, loop through the location list and make up the srvlist. */ xdrsum = (2 * NFSX_UNSIGNED) + NFSM_RNDUP(len); cp2 = cp3 = malloc(1024, M_NFSSTRING, M_WAITOK); slen = 1024; siz = 0; for (i = 0; i < cnt; i++) { SLIST_INIT(&head); NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); nsrv = fxdr_unsigned(int, *tl); if (nsrv <= 0) { error = NFSERR_BADXDR; goto nfsmout; } /* * Handle the first server by putting it in the srvstr. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len <= 0 || len > 1024) { error = NFSERR_BADXDR; goto nfsmout; } nfsrv_refstrbigenough(siz + len + 3, &cp2, &cp3, &slen); if (cp3 != cp2) { *cp3++ = ','; siz++; } error = nfsrv_mtostr(nd, cp3, len); if (error) goto nfsmout; cp3 += len; *cp3++ = ':'; siz += (len + 1); xdrsum += (2 * NFSX_UNSIGNED) + NFSM_RNDUP(len); for (j = 1; j < nsrv; j++) { /* * Yuck, put them in an slist and process them later. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len <= 0 || len > 1024) { error = NFSERR_BADXDR; goto nfsmout; } lsp = (struct list *)malloc(sizeof (struct list) + len, M_TEMP, M_WAITOK); error = nfsrv_mtostr(nd, lsp->host, len); if (error) goto nfsmout; xdrsum += NFSX_UNSIGNED + NFSM_RNDUP(len); lsp->len = len; SLIST_INSERT_HEAD(&head, lsp, next); } /* * Finally, we can get the path. */ NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); len = fxdr_unsigned(int, *tl); if (len <= 0 || len > 1024) { error = NFSERR_BADXDR; goto nfsmout; } nfsrv_refstrbigenough(siz + len + 1, &cp2, &cp3, &slen); error = nfsrv_mtostr(nd, cp3, len); if (error) goto nfsmout; xdrsum += NFSX_UNSIGNED + NFSM_RNDUP(len); str = cp3; stringlen = len; cp3 += len; siz += len; SLIST_FOREACH_SAFE(lsp, &head, next, nlsp) { nfsrv_refstrbigenough(siz + lsp->len + stringlen + 3, &cp2, &cp3, &slen); *cp3++ = ','; NFSBCOPY(lsp->host, cp3, lsp->len); cp3 += lsp->len; *cp3++ = ':'; NFSBCOPY(str, cp3, stringlen); cp3 += stringlen; *cp3 = '\0'; siz += (lsp->len + stringlen + 2); free((caddr_t)lsp, M_TEMP); } } *fsrootp = cp; *srvp = cp2; *sump = xdrsum; NFSEXITCODE2(0, nd); return (0); nfsmout: if (cp != NULL) free(cp, M_NFSSTRING); if (cp2 != NULL) free(cp2, M_NFSSTRING); NFSEXITCODE2(error, nd); return (error); } /* * Make the malloc'd space large enough. This is a pain, but the xdr * doesn't set an upper bound on the side, so... */ static void nfsrv_refstrbigenough(int siz, u_char **cpp, u_char **cpp2, int *slenp) { u_char *cp; int i; if (siz <= *slenp) return; cp = malloc(siz + 1024, M_NFSSTRING, M_WAITOK); NFSBCOPY(*cpp, cp, *slenp); free(*cpp, M_NFSSTRING); i = *cpp2 - *cpp; *cpp = cp; *cpp2 = cp + i; *slenp = siz + 1024; } /* * Initialize the reply header data structures. */ APPLESTATIC void nfsrvd_rephead(struct nfsrv_descript *nd) { mbuf_t mreq; /* * If this is a big reply, use a cluster. */ if ((nd->nd_flag & ND_GSSINITREPLY) == 0 && nfs_bigreply[nd->nd_procnum]) { NFSMCLGET(mreq, M_WAITOK); nd->nd_mreq = mreq; nd->nd_mb = mreq; } else { NFSMGET(mreq); nd->nd_mreq = mreq; nd->nd_mb = mreq; } nd->nd_bpos = NFSMTOD(mreq, caddr_t); mbuf_setlen(mreq, 0); if ((nd->nd_flag & ND_GSSINITREPLY) == 0) NFSM_BUILD(nd->nd_errp, int *, NFSX_UNSIGNED); } /* * Lock a socket against others. * Currently used to serialize connect/disconnect attempts. */ int newnfs_sndlock(int *flagp) { struct timespec ts; NFSLOCKSOCK(); while (*flagp & NFSR_SNDLOCK) { *flagp |= NFSR_WANTSND; ts.tv_sec = 0; ts.tv_nsec = 0; (void) nfsmsleep((caddr_t)flagp, NFSSOCKMUTEXPTR, PZERO - 1, "nfsndlck", &ts); } *flagp |= NFSR_SNDLOCK; NFSUNLOCKSOCK(); return (0); } /* * Unlock the stream socket for others. */ void newnfs_sndunlock(int *flagp) { NFSLOCKSOCK(); if ((*flagp & NFSR_SNDLOCK) == 0) panic("nfs sndunlock"); *flagp &= ~NFSR_SNDLOCK; if (*flagp & NFSR_WANTSND) { *flagp &= ~NFSR_WANTSND; wakeup((caddr_t)flagp); } NFSUNLOCKSOCK(); } APPLESTATIC int nfsv4_getipaddr(struct nfsrv_descript *nd, struct sockaddr_storage *sa, int *isudp) { struct sockaddr_in *sad; struct sockaddr_in6 *sad6; struct in_addr saddr; uint32_t portnum, *tl; int af = 0, i, j, k; char addr[64], protocol[5], *cp; int cantparse = 0, error = 0; uint16_t portv; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); if (i >= 3 && i <= 4) { error = nfsrv_mtostr(nd, protocol, i); if (error) goto nfsmout; if (strcmp(protocol, "tcp") == 0) { af = AF_INET; *isudp = 0; } else if (strcmp(protocol, "udp") == 0) { af = AF_INET; *isudp = 1; } else if (strcmp(protocol, "tcp6") == 0) { af = AF_INET6; *isudp = 0; } else if (strcmp(protocol, "udp6") == 0) { af = AF_INET6; *isudp = 1; } else cantparse = 1; } else { cantparse = 1; if (i > 0) { error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; } } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); i = fxdr_unsigned(int, *tl); if (i < 0) { error = NFSERR_BADXDR; goto nfsmout; } else if (cantparse == 0 && i >= 11 && i < 64) { /* * The shortest address is 11chars and the longest is < 64. */ error = nfsrv_mtostr(nd, addr, i); if (error) goto nfsmout; /* Find the port# at the end and extract that. */ i = strlen(addr); k = 0; cp = &addr[i - 1]; /* Count back two '.'s from end to get port# field. */ for (j = 0; j < i; j++) { if (*cp == '.') { k++; if (k == 2) break; } cp--; } if (k == 2) { /* * The NFSv4 port# is appended as .N.N, where N is * a decimal # in the range 0-255, just like an inet4 * address. Cheat and use inet_aton(), which will * return a Class A address and then shift the high * order 8bits over to convert it to the port#. */ *cp++ = '\0'; if (inet_aton(cp, &saddr) == 1) { portnum = ntohl(saddr.s_addr); portv = (uint16_t)((portnum >> 16) | (portnum & 0xff)); } else cantparse = 1; } else cantparse = 1; if (cantparse == 0) { if (af == AF_INET) { sad = (struct sockaddr_in *)sa; if (inet_pton(af, addr, &sad->sin_addr) == 1) { sad->sin_len = sizeof(*sad); sad->sin_family = AF_INET; sad->sin_port = htons(portv); return (0); } } else { sad6 = (struct sockaddr_in6 *)sa; if (inet_pton(af, addr, &sad6->sin6_addr) == 1) { sad6->sin6_len = sizeof(*sad6); sad6->sin6_family = AF_INET6; sad6->sin6_port = htons(portv); return (0); } } } } else { if (i > 0) { error = nfsm_advance(nd, NFSM_RNDUP(i), -1); if (error) goto nfsmout; } } error = EPERM; nfsmout: return (error); } /* * Handle an NFSv4.1 Sequence request for the session. * If reply != NULL, use it to return the cached reply, as required. * The client gets a cached reply via this call for callbacks, however the * server gets a cached reply via the nfsv4_seqsess_cachereply() call. */ int nfsv4_seqsession(uint32_t seqid, uint32_t slotid, uint32_t highslot, struct nfsslot *slots, struct mbuf **reply, uint16_t maxslot) { int error; error = 0; if (reply != NULL) *reply = NULL; if (slotid > maxslot) return (NFSERR_BADSLOT); if (seqid == slots[slotid].nfssl_seq) { /* A retry. */ if (slots[slotid].nfssl_inprog != 0) error = NFSERR_DELAY; else if (slots[slotid].nfssl_reply != NULL) { if (reply != NULL) { *reply = slots[slotid].nfssl_reply; slots[slotid].nfssl_reply = NULL; } slots[slotid].nfssl_inprog = 1; error = NFSERR_REPLYFROMCACHE; } else /* No reply cached, so just do it. */ slots[slotid].nfssl_inprog = 1; } else if ((slots[slotid].nfssl_seq + 1) == seqid) { if (slots[slotid].nfssl_reply != NULL) m_freem(slots[slotid].nfssl_reply); slots[slotid].nfssl_reply = NULL; slots[slotid].nfssl_inprog = 1; slots[slotid].nfssl_seq++; } else error = NFSERR_SEQMISORDERED; return (error); } /* * Cache this reply for the slot. * Use the "rep" argument to return the cached reply if repstat is set to * NFSERR_REPLYFROMCACHE. The client never sets repstat to this value. */ void nfsv4_seqsess_cacherep(uint32_t slotid, struct nfsslot *slots, int repstat, struct mbuf **rep) { if (repstat == NFSERR_REPLYFROMCACHE) { *rep = slots[slotid].nfssl_reply; slots[slotid].nfssl_reply = NULL; } else { if (slots[slotid].nfssl_reply != NULL) m_freem(slots[slotid].nfssl_reply); slots[slotid].nfssl_reply = *rep; } slots[slotid].nfssl_inprog = 0; } /* * Generate the xdr for an NFSv4.1 Sequence Operation. */ APPLESTATIC void nfsv4_setsequence(struct nfsmount *nmp, struct nfsrv_descript *nd, struct nfsclsession *sep, int dont_replycache) { uint32_t *tl, slotseq = 0; int error, maxslot, slotpos; uint8_t sessionid[NFSX_V4SESSIONID]; error = nfsv4_sequencelookup(nmp, sep, &slotpos, &maxslot, &slotseq, sessionid); /* Build the Sequence arguments. */ NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED); nd->nd_sequence = tl; bcopy(sessionid, tl, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; nd->nd_slotseq = tl; if (error == 0) { *tl++ = txdr_unsigned(slotseq); *tl++ = txdr_unsigned(slotpos); *tl++ = txdr_unsigned(maxslot); if (dont_replycache == 0) *tl = newnfs_true; else *tl = newnfs_false; } else { /* * There are two errors and the rest of the session can * just be zeros. * NFSERR_BADSESSION: This bad session should just generate * the same error again when the RPC is retried. * ESTALE: A forced dismount is in progress and will cause the * RPC to fail later. */ *tl++ = 0; *tl++ = 0; *tl++ = 0; *tl = 0; } nd->nd_flag |= ND_HASSEQUENCE; } int nfsv4_sequencelookup(struct nfsmount *nmp, struct nfsclsession *sep, int *slotposp, int *maxslotp, uint32_t *slotseqp, uint8_t *sessionid) { int i, maxslot, slotpos; uint64_t bitval; /* Find an unused slot. */ slotpos = -1; maxslot = -1; mtx_lock(&sep->nfsess_mtx); do { if (nmp != NULL && sep->nfsess_defunct != 0) { /* Just return the bad session. */ bcopy(sep->nfsess_sessionid, sessionid, NFSX_V4SESSIONID); mtx_unlock(&sep->nfsess_mtx); return (NFSERR_BADSESSION); } bitval = 1; for (i = 0; i < sep->nfsess_foreslots; i++) { if ((bitval & sep->nfsess_slots) == 0) { slotpos = i; sep->nfsess_slots |= bitval; sep->nfsess_slotseq[i]++; *slotseqp = sep->nfsess_slotseq[i]; break; } bitval <<= 1; } if (slotpos == -1) { /* * If a forced dismount is in progress, just return. * This RPC attempt will fail when it calls * newnfs_request(). */ - if (nmp != NULL && - (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) - != 0) { + if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) { mtx_unlock(&sep->nfsess_mtx); return (ESTALE); } /* Wake up once/sec, to check for a forced dismount. */ (void)mtx_sleep(&sep->nfsess_slots, &sep->nfsess_mtx, PZERO, "nfsclseq", hz); } } while (slotpos == -1); /* Now, find the highest slot in use. (nfsc_slots is 64bits) */ bitval = 1; for (i = 0; i < 64; i++) { if ((bitval & sep->nfsess_slots) != 0) maxslot = i; bitval <<= 1; } bcopy(sep->nfsess_sessionid, sessionid, NFSX_V4SESSIONID); mtx_unlock(&sep->nfsess_mtx); *slotposp = slotpos; *maxslotp = maxslot; return (0); } /* * Free a session slot. */ APPLESTATIC void nfsv4_freeslot(struct nfsclsession *sep, int slot) { uint64_t bitval; bitval = 1; if (slot > 0) bitval <<= slot; mtx_lock(&sep->nfsess_mtx); if ((bitval & sep->nfsess_slots) == 0) printf("freeing free slot!!\n"); sep->nfsess_slots &= ~bitval; wakeup(&sep->nfsess_slots); mtx_unlock(&sep->nfsess_mtx); } Index: head/sys/fs/nfs/nfscl.h =================================================================== --- head/sys/fs/nfs/nfscl.h (revision 321627) +++ head/sys/fs/nfs/nfscl.h (revision 321628) @@ -1,77 +1,80 @@ /*- * Copyright (c) 2009 Rick Macklem, University of Guelph * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _NFS_NFSCL_H #define _NFS_NFSCL_H /* * Extra stuff for a NFSv4 nfsnode. * MALLOC'd to the correct length for the name and file handle. * n4_data has the file handle, followed by the file name. * The macro NFS4NODENAME() returns a pointer to the start of the * name. */ struct nfsv4node { u_int16_t n4_fhlen; u_int16_t n4_namelen; u_int8_t n4_data[1]; }; #define NFS4NODENAME(n) (&((n)->n4_data[(n)->n4_fhlen])) /* * Just a macro to convert the nfscl_reqstart arguments. */ #define NFSCL_REQSTART(n, p, v) \ nfscl_reqstart((n), (p), VFSTONFS((v)->v_mount), \ VTONFS(v)->n_fhp->nfh_fh, VTONFS(v)->n_fhp->nfh_len, NULL, NULL) /* * These two macros convert between a lease duration and renew interval. * For now, just make the renew interval 1/2 the lease duration. * (They should be inverse operators.) */ #define NFSCL_RENEW(l) (((l) < 2) ? 1 : ((l) / 2)) #define NFSCL_LEASE(r) ((r) * 2) +/* This macro checks to see if a forced dismount is about to occur. */ +#define NFSCL_FORCEDISM(m) (((m)->mnt_kern_flag & MNTK_UNMOUNTF) != 0) + /* * These flag bits are used for the argument to nfscl_fillsattr() to * indicate special handling of the attributes. */ #define NFSSATTR_FULL 0x1 #define NFSSATTR_SIZE0 0x2 #define NFSSATTR_SIZENEG1 0x4 #define NFSSATTR_SIZERDEV 0x8 /* Use this macro for debug printfs. */ #define NFSCL_DEBUG(level, ...) do { \ if (nfscl_debuglevel >= (level)) \ printf(__VA_ARGS__); \ } while (0) #endif /* _NFS_NFSCL_H */ Index: head/sys/fs/nfsclient/nfs_clbio.c =================================================================== --- head/sys/fs/nfsclient/nfs_clbio.c (revision 321627) +++ head/sys/fs/nfsclient/nfs_clbio.c (revision 321628) @@ -1,1897 +1,1897 @@ /*- * 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. * * @(#)nfs_bio.c 8.9 (Berkeley) 3/30/95 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include extern int newnfs_directio_allow_mmap; extern struct nfsstatsv1 nfsstatsv1; extern struct mtx ncl_iod_mutex; extern int ncl_numasync; extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON]; extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON]; extern int newnfs_directio_enable; extern int nfs_keep_dirty_on_error; int ncl_pbuf_freecnt = -1; /* start out unlimited */ static struct buf *nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, struct thread *td); static int nfs_directio_write(struct vnode *vp, struct uio *uiop, struct ucred *cred, int ioflag); /* * Vnode op for VM getpages. */ SYSCTL_DECL(_vfs_nfs); static int use_buf_pager = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0, "Use buffer pager instead of direct readrpc call"); static daddr_t ncl_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) { return (off / vp->v_bufobj.bo_bsize); } static int ncl_gbp_getblksz(struct vnode *vp, daddr_t lbn) { struct nfsnode *np; u_quad_t nsize; int biosize, bcount; np = VTONFS(vp); mtx_lock(&np->n_mtx); nsize = np->n_size; mtx_unlock(&np->n_mtx); biosize = vp->v_bufobj.bo_bsize; bcount = biosize; if ((off_t)lbn * biosize >= nsize) bcount = 0; else if ((off_t)(lbn + 1) * biosize > nsize) bcount = nsize - (off_t)lbn * biosize; return (bcount); } int ncl_getpages(struct vop_getpages_args *ap) { int i, error, nextoff, size, toff, count, npages; struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; struct vnode *vp; struct thread *td; struct ucred *cred; struct nfsmount *nmp; vm_object_t object; vm_page_t *pages; struct nfsnode *np; vp = ap->a_vp; np = VTONFS(vp); td = curthread; cred = curthread->td_ucred; nmp = VFSTONFS(vp->v_mount); pages = ap->a_m; npages = ap->a_count; if ((object = vp->v_object) == NULL) { printf("ncl_getpages: called with non-merged cache vnode\n"); return (VM_PAGER_ERROR); } if (newnfs_directio_enable && !newnfs_directio_allow_mmap) { mtx_lock(&np->n_mtx); if ((np->n_flag & NNONCACHE) && (vp->v_type == VREG)) { mtx_unlock(&np->n_mtx); printf("ncl_getpages: called on non-cacheable vnode\n"); return (VM_PAGER_ERROR); } else mtx_unlock(&np->n_mtx); } mtx_lock(&nmp->nm_mtx); if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { mtx_unlock(&nmp->nm_mtx); /* We'll never get here for v4, because we always have fsinfo */ (void)ncl_fsinfo(nmp, vp, cred, td); } else mtx_unlock(&nmp->nm_mtx); if (use_buf_pager) return (vfs_bio_getpages(vp, pages, npages, ap->a_rbehind, ap->a_rahead, ncl_gbp_getblkno, ncl_gbp_getblksz)); /* * If the requested page is partially valid, just return it and * allow the pager to zero-out the blanks. Partially valid pages * can only occur at the file EOF. * * XXXGL: is that true for NFS, where short read can occur??? */ VM_OBJECT_WLOCK(object); if (pages[npages - 1]->valid != 0 && --npages == 0) goto out; VM_OBJECT_WUNLOCK(object); /* * We use only the kva address for the buffer, but this is extremely * convenient and fast. */ bp = getpbuf(&ncl_pbuf_freecnt); kva = (vm_offset_t) bp->b_data; pmap_qenter(kva, pages, npages); VM_CNT_INC(v_vnodein); VM_CNT_ADD(v_vnodepgsin, npages); count = npages << PAGE_SHIFT; iov.iov_base = (caddr_t) kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = IDX_TO_OFF(pages[0]->pindex); uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = td; error = ncl_readrpc(vp, &uio, cred); pmap_qremove(kva, npages); relpbuf(bp, &ncl_pbuf_freecnt); if (error && (uio.uio_resid == count)) { printf("ncl_getpages: error %d\n", error); return (VM_PAGER_ERROR); } /* * Calculate the number of bytes read and validate only that number * of bytes. Note that due to pending writes, size may be 0. This * does not mean that the remaining data is invalid! */ size = count - uio.uio_resid; VM_OBJECT_WLOCK(object); for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { vm_page_t m; nextoff = toff + PAGE_SIZE; m = pages[i]; if (nextoff <= size) { /* * Read operation filled an entire page */ m->valid = VM_PAGE_BITS_ALL; KASSERT(m->dirty == 0, ("nfs_getpages: page %p is dirty", m)); } else if (size > toff) { /* * Read operation filled a partial page. */ m->valid = 0; vm_page_set_valid_range(m, 0, size - toff); KASSERT(m->dirty == 0, ("nfs_getpages: page %p is dirty", m)); } else { /* * Read operation was short. If no error * occurred we may have hit a zero-fill * section. We leave valid set to 0, and page * is freed by vm_page_readahead_finish() if * its index is not equal to requested, or * page is zeroed and set valid by * vm_pager_get_pages() for requested page. */ ; } } out: VM_OBJECT_WUNLOCK(object); if (ap->a_rbehind) *ap->a_rbehind = 0; if (ap->a_rahead) *ap->a_rahead = 0; return (VM_PAGER_OK); } /* * Vnode op for VM putpages. */ int ncl_putpages(struct vop_putpages_args *ap) { struct uio uio; struct iovec iov; int i, error, npages, count; off_t offset; int *rtvals; struct vnode *vp; struct thread *td; struct ucred *cred; struct nfsmount *nmp; struct nfsnode *np; vm_page_t *pages; vp = ap->a_vp; np = VTONFS(vp); td = curthread; /* XXX */ /* Set the cred to n_writecred for the write rpcs. */ if (np->n_writecred != NULL) cred = crhold(np->n_writecred); else cred = crhold(curthread->td_ucred); /* XXX */ nmp = VFSTONFS(vp->v_mount); pages = ap->a_m; count = ap->a_count; rtvals = ap->a_rtvals; npages = btoc(count); offset = IDX_TO_OFF(pages[0]->pindex); mtx_lock(&nmp->nm_mtx); if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { mtx_unlock(&nmp->nm_mtx); (void)ncl_fsinfo(nmp, vp, cred, td); } else mtx_unlock(&nmp->nm_mtx); mtx_lock(&np->n_mtx); if (newnfs_directio_enable && !newnfs_directio_allow_mmap && (np->n_flag & NNONCACHE) && (vp->v_type == VREG)) { mtx_unlock(&np->n_mtx); printf("ncl_putpages: called on noncache-able vnode\n"); mtx_lock(&np->n_mtx); } /* * When putting pages, do not extend file past EOF. */ if (offset + count > np->n_size) { count = np->n_size - offset; if (count < 0) count = 0; } mtx_unlock(&np->n_mtx); for (i = 0; i < npages; i++) rtvals[i] = VM_PAGER_ERROR; VM_CNT_INC(v_vnodeout); VM_CNT_ADD(v_vnodepgsout, count); iov.iov_base = unmapped_buf; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = offset; uio.uio_resid = count; uio.uio_segflg = UIO_NOCOPY; uio.uio_rw = UIO_WRITE; uio.uio_td = td; error = VOP_WRITE(vp, &uio, vnode_pager_putpages_ioflags(ap->a_sync), cred); crfree(cred); if (error == 0 || !nfs_keep_dirty_on_error) { vnode_pager_undirty_pages(pages, rtvals, count - uio.uio_resid, np->n_size - offset, npages * PAGE_SIZE); } return (rtvals[0]); } /* * For nfs, cache consistency can only be maintained approximately. * Although RFC1094 does not specify the criteria, the following is * believed to be compatible with the reference port. * For nfs: * If the file's modify time on the server has changed since the * last read rpc or you have written to the file, * you may have lost data cache consistency with the * server, so flush all of the file's data out of the cache. * Then force a getattr rpc to ensure that you have up to date * attributes. * NB: This implies that cache data can be read when up to * NFS_ATTRTIMEO seconds out of date. If you find that you need current * attributes this could be forced by setting n_attrstamp to 0 before * the VOP_GETATTR() call. */ static inline int nfs_bioread_check_cons(struct vnode *vp, struct thread *td, struct ucred *cred) { int error = 0; struct vattr vattr; struct nfsnode *np = VTONFS(vp); int old_lock; /* * Grab the exclusive lock before checking whether the cache is * consistent. * XXX - We can make this cheaper later (by acquiring cheaper locks). * But for now, this suffices. */ old_lock = ncl_upgrade_vnlock(vp); if (vp->v_iflag & VI_DOOMED) { error = EBADF; goto out; } mtx_lock(&np->n_mtx); if (np->n_flag & NMODIFIED) { mtx_unlock(&np->n_mtx); if (vp->v_type != VREG) { if (vp->v_type != VDIR) panic("nfs: bioread, not dir"); ncl_invaldir(vp); error = ncl_vinvalbuf(vp, V_SAVE, td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) error = EBADF; if (error != 0) goto out; } np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); error = VOP_GETATTR(vp, &vattr, cred); if (error) goto out; mtx_lock(&np->n_mtx); np->n_mtime = vattr.va_mtime; mtx_unlock(&np->n_mtx); } else { mtx_unlock(&np->n_mtx); error = VOP_GETATTR(vp, &vattr, cred); if (error) return (error); mtx_lock(&np->n_mtx); if ((np->n_flag & NSIZECHANGED) || (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime))) { mtx_unlock(&np->n_mtx); if (vp->v_type == VDIR) ncl_invaldir(vp); error = ncl_vinvalbuf(vp, V_SAVE, td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) error = EBADF; if (error != 0) goto out; mtx_lock(&np->n_mtx); np->n_mtime = vattr.va_mtime; np->n_flag &= ~NSIZECHANGED; } mtx_unlock(&np->n_mtx); } out: ncl_downgrade_vnlock(vp, old_lock); return (error); } /* * Vnode op for read using bio */ int ncl_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred) { struct nfsnode *np = VTONFS(vp); int biosize, i; struct buf *bp, *rabp; struct thread *td; struct nfsmount *nmp = VFSTONFS(vp->v_mount); daddr_t lbn, rabn; int bcount; int seqcount; int nra, error = 0, n = 0, on = 0; off_t tmp_off; KASSERT(uio->uio_rw == UIO_READ, ("ncl_read mode")); if (uio->uio_resid == 0) return (0); if (uio->uio_offset < 0) /* XXX VDIR cookies can be negative */ return (EINVAL); td = uio->uio_td; mtx_lock(&nmp->nm_mtx); if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { mtx_unlock(&nmp->nm_mtx); (void)ncl_fsinfo(nmp, vp, cred, td); mtx_lock(&nmp->nm_mtx); } if (nmp->nm_rsize == 0 || nmp->nm_readdirsize == 0) (void) newnfs_iosize(nmp); tmp_off = uio->uio_offset + uio->uio_resid; if (vp->v_type != VDIR && (tmp_off > nmp->nm_maxfilesize || tmp_off < uio->uio_offset)) { mtx_unlock(&nmp->nm_mtx); return (EFBIG); } mtx_unlock(&nmp->nm_mtx); if (newnfs_directio_enable && (ioflag & IO_DIRECT) && (vp->v_type == VREG)) /* No caching/ no readaheads. Just read data into the user buffer */ return ncl_readrpc(vp, uio, cred); biosize = vp->v_bufobj.bo_bsize; seqcount = (int)((off_t)(ioflag >> IO_SEQSHIFT) * biosize / BKVASIZE); error = nfs_bioread_check_cons(vp, td, cred); if (error) return error; do { u_quad_t nsize; mtx_lock(&np->n_mtx); nsize = np->n_size; mtx_unlock(&np->n_mtx); switch (vp->v_type) { case VREG: NFSINCRGLOBAL(nfsstatsv1.biocache_reads); lbn = uio->uio_offset / biosize; on = uio->uio_offset - (lbn * biosize); /* * Start the read ahead(s), as required. */ if (nmp->nm_readahead > 0) { for (nra = 0; nra < nmp->nm_readahead && nra < seqcount && (off_t)(lbn + 1 + nra) * biosize < nsize; nra++) { rabn = lbn + 1 + nra; if (incore(&vp->v_bufobj, rabn) == NULL) { rabp = nfs_getcacheblk(vp, rabn, biosize, td); if (!rabp) { error = newnfs_sigintr(nmp, td); return (error ? error : EINTR); } if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { rabp->b_flags |= B_ASYNC; rabp->b_iocmd = BIO_READ; vfs_busy_pages(rabp, 0); if (ncl_asyncio(nmp, rabp, cred, td)) { rabp->b_flags |= B_INVAL; rabp->b_ioflags |= BIO_ERROR; vfs_unbusy_pages(rabp); brelse(rabp); break; } } else { brelse(rabp); } } } } /* Note that bcount is *not* DEV_BSIZE aligned. */ bcount = biosize; if ((off_t)lbn * biosize >= nsize) { bcount = 0; } else if ((off_t)(lbn + 1) * biosize > nsize) { bcount = nsize - (off_t)lbn * biosize; } bp = nfs_getcacheblk(vp, lbn, bcount, td); if (!bp) { error = newnfs_sigintr(nmp, td); return (error ? error : EINTR); } /* * If B_CACHE is not set, we must issue the read. If this * fails, we return an error. */ if ((bp->b_flags & B_CACHE) == 0) { bp->b_iocmd = BIO_READ; vfs_busy_pages(bp, 0); error = ncl_doio(vp, bp, cred, td, 0); if (error) { brelse(bp); return (error); } } /* * on is the offset into the current bp. Figure out how many * bytes we can copy out of the bp. Note that bcount is * NOT DEV_BSIZE aligned. * * Then figure out how many bytes we can copy into the uio. */ n = 0; if (on < bcount) n = MIN((unsigned)(bcount - on), uio->uio_resid); break; case VLNK: NFSINCRGLOBAL(nfsstatsv1.biocache_readlinks); bp = nfs_getcacheblk(vp, (daddr_t)0, NFS_MAXPATHLEN, td); if (!bp) { error = newnfs_sigintr(nmp, td); return (error ? error : EINTR); } if ((bp->b_flags & B_CACHE) == 0) { bp->b_iocmd = BIO_READ; vfs_busy_pages(bp, 0); error = ncl_doio(vp, bp, cred, td, 0); if (error) { bp->b_ioflags |= BIO_ERROR; brelse(bp); return (error); } } n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid); on = 0; break; case VDIR: NFSINCRGLOBAL(nfsstatsv1.biocache_readdirs); if (np->n_direofoffset && uio->uio_offset >= np->n_direofoffset) { return (0); } lbn = (uoff_t)uio->uio_offset / NFS_DIRBLKSIZ; on = uio->uio_offset & (NFS_DIRBLKSIZ - 1); bp = nfs_getcacheblk(vp, lbn, NFS_DIRBLKSIZ, td); if (!bp) { error = newnfs_sigintr(nmp, td); return (error ? error : EINTR); } if ((bp->b_flags & B_CACHE) == 0) { bp->b_iocmd = BIO_READ; vfs_busy_pages(bp, 0); error = ncl_doio(vp, bp, cred, td, 0); if (error) { brelse(bp); } while (error == NFSERR_BAD_COOKIE) { ncl_invaldir(vp); error = ncl_vinvalbuf(vp, 0, td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) return (EBADF); /* * Yuck! The directory has been modified on the * server. The only way to get the block is by * reading from the beginning to get all the * offset cookies. * * Leave the last bp intact unless there is an error. * Loop back up to the while if the error is another * NFSERR_BAD_COOKIE (double yuch!). */ for (i = 0; i <= lbn && !error; i++) { if (np->n_direofoffset && (i * NFS_DIRBLKSIZ) >= np->n_direofoffset) return (0); bp = nfs_getcacheblk(vp, i, NFS_DIRBLKSIZ, td); if (!bp) { error = newnfs_sigintr(nmp, td); return (error ? error : EINTR); } if ((bp->b_flags & B_CACHE) == 0) { bp->b_iocmd = BIO_READ; vfs_busy_pages(bp, 0); error = ncl_doio(vp, bp, cred, td, 0); /* * no error + B_INVAL == directory EOF, * use the block. */ if (error == 0 && (bp->b_flags & B_INVAL)) break; } /* * An error will throw away the block and the * for loop will break out. If no error and this * is not the block we want, we throw away the * block and go for the next one via the for loop. */ if (error || i < lbn) brelse(bp); } } /* * The above while is repeated if we hit another cookie * error. If we hit an error and it wasn't a cookie error, * we give up. */ if (error) return (error); } /* * If not eof and read aheads are enabled, start one. * (You need the current block first, so that you have the * directory offset cookie of the next block.) */ if (nmp->nm_readahead > 0 && (bp->b_flags & B_INVAL) == 0 && (np->n_direofoffset == 0 || (lbn + 1) * NFS_DIRBLKSIZ < np->n_direofoffset) && incore(&vp->v_bufobj, lbn + 1) == NULL) { rabp = nfs_getcacheblk(vp, lbn + 1, NFS_DIRBLKSIZ, td); if (rabp) { if ((rabp->b_flags & (B_CACHE|B_DELWRI)) == 0) { rabp->b_flags |= B_ASYNC; rabp->b_iocmd = BIO_READ; vfs_busy_pages(rabp, 0); if (ncl_asyncio(nmp, rabp, cred, td)) { rabp->b_flags |= B_INVAL; rabp->b_ioflags |= BIO_ERROR; vfs_unbusy_pages(rabp); brelse(rabp); } } else { brelse(rabp); } } } /* * Unlike VREG files, whos buffer size ( bp->b_bcount ) is * chopped for the EOF condition, we cannot tell how large * NFS directories are going to be until we hit EOF. So * an NFS directory buffer is *not* chopped to its EOF. Now, * it just so happens that b_resid will effectively chop it * to EOF. *BUT* this information is lost if the buffer goes * away and is reconstituted into a B_CACHE state ( due to * being VMIO ) later. So we keep track of the directory eof * in np->n_direofoffset and chop it off as an extra step * right here. */ n = lmin(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid - on); if (np->n_direofoffset && n > np->n_direofoffset - uio->uio_offset) n = np->n_direofoffset - uio->uio_offset; break; default: printf(" ncl_bioread: type %x unexpected\n", vp->v_type); bp = NULL; break; } if (n > 0) { error = vn_io_fault_uiomove(bp->b_data + on, (int)n, uio); } if (vp->v_type == VLNK) n = 0; if (bp != NULL) brelse(bp); } while (error == 0 && uio->uio_resid > 0 && n > 0); return (error); } /* * The NFS write path cannot handle iovecs with len > 1. So we need to * break up iovecs accordingly (restricting them to wsize). * For the SYNC case, we can do this with 1 copy (user buffer -> mbuf). * For the ASYNC case, 2 copies are needed. The first a copy from the * user buffer to a staging buffer and then a second copy from the staging * buffer to mbufs. This can be optimized by copying from the user buffer * directly into mbufs and passing the chain down, but that requires a * fair amount of re-working of the relevant codepaths (and can be done * later). */ static int nfs_directio_write(vp, uiop, cred, ioflag) struct vnode *vp; struct uio *uiop; struct ucred *cred; int ioflag; { int error; struct nfsmount *nmp = VFSTONFS(vp->v_mount); struct thread *td = uiop->uio_td; int size; int wsize; mtx_lock(&nmp->nm_mtx); wsize = nmp->nm_wsize; mtx_unlock(&nmp->nm_mtx); if (ioflag & IO_SYNC) { int iomode, must_commit; struct uio uio; struct iovec iov; do_sync: while (uiop->uio_resid > 0) { size = MIN(uiop->uio_resid, wsize); size = MIN(uiop->uio_iov->iov_len, size); iov.iov_base = uiop->uio_iov->iov_base; iov.iov_len = size; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = uiop->uio_offset; uio.uio_resid = size; uio.uio_segflg = UIO_USERSPACE; uio.uio_rw = UIO_WRITE; uio.uio_td = td; iomode = NFSWRITE_FILESYNC; error = ncl_writerpc(vp, &uio, cred, &iomode, &must_commit, 0); KASSERT((must_commit == 0), ("ncl_directio_write: Did not commit write")); if (error) return (error); uiop->uio_offset += size; uiop->uio_resid -= size; if (uiop->uio_iov->iov_len <= size) { uiop->uio_iovcnt--; uiop->uio_iov++; } else { uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + size; uiop->uio_iov->iov_len -= size; } } } else { struct uio *t_uio; struct iovec *t_iov; struct buf *bp; /* * Break up the write into blocksize chunks and hand these * over to nfsiod's for write back. * Unfortunately, this incurs a copy of the data. Since * the user could modify the buffer before the write is * initiated. * * The obvious optimization here is that one of the 2 copies * in the async write path can be eliminated by copying the * data here directly into mbufs and passing the mbuf chain * down. But that will require a fair amount of re-working * of the code and can be done if there's enough interest * in NFS directio access. */ while (uiop->uio_resid > 0) { size = MIN(uiop->uio_resid, wsize); size = MIN(uiop->uio_iov->iov_len, size); bp = getpbuf(&ncl_pbuf_freecnt); t_uio = malloc(sizeof(struct uio), M_NFSDIRECTIO, M_WAITOK); t_iov = malloc(sizeof(struct iovec), M_NFSDIRECTIO, M_WAITOK); t_iov->iov_base = malloc(size, M_NFSDIRECTIO, M_WAITOK); t_iov->iov_len = size; t_uio->uio_iov = t_iov; t_uio->uio_iovcnt = 1; t_uio->uio_offset = uiop->uio_offset; t_uio->uio_resid = size; t_uio->uio_segflg = UIO_SYSSPACE; t_uio->uio_rw = UIO_WRITE; t_uio->uio_td = td; KASSERT(uiop->uio_segflg == UIO_USERSPACE || uiop->uio_segflg == UIO_SYSSPACE, ("nfs_directio_write: Bad uio_segflg")); if (uiop->uio_segflg == UIO_USERSPACE) { error = copyin(uiop->uio_iov->iov_base, t_iov->iov_base, size); if (error != 0) goto err_free; } else /* * UIO_SYSSPACE may never happen, but handle * it just in case it does. */ bcopy(uiop->uio_iov->iov_base, t_iov->iov_base, size); bp->b_flags |= B_DIRECT; bp->b_iocmd = BIO_WRITE; if (cred != NOCRED) { crhold(cred); bp->b_wcred = cred; } else bp->b_wcred = NOCRED; bp->b_caller1 = (void *)t_uio; bp->b_vp = vp; error = ncl_asyncio(nmp, bp, NOCRED, td); err_free: if (error) { free(t_iov->iov_base, M_NFSDIRECTIO); free(t_iov, M_NFSDIRECTIO); free(t_uio, M_NFSDIRECTIO); bp->b_vp = NULL; relpbuf(bp, &ncl_pbuf_freecnt); if (error == EINTR) return (error); goto do_sync; } uiop->uio_offset += size; uiop->uio_resid -= size; if (uiop->uio_iov->iov_len <= size) { uiop->uio_iovcnt--; uiop->uio_iov++; } else { uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base + size; uiop->uio_iov->iov_len -= size; } } } return (0); } /* * Vnode op for write using bio */ int ncl_write(struct vop_write_args *ap) { int biosize; struct uio *uio = ap->a_uio; struct thread *td = uio->uio_td; struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct ucred *cred = ap->a_cred; int ioflag = ap->a_ioflag; struct buf *bp; struct vattr vattr; struct nfsmount *nmp = VFSTONFS(vp->v_mount); daddr_t lbn; int bcount, noncontig_write, obcount; int bp_cached, n, on, error = 0, error1, wouldcommit; size_t orig_resid, local_resid; off_t orig_size, tmp_off; KASSERT(uio->uio_rw == UIO_WRITE, ("ncl_write mode")); KASSERT(uio->uio_segflg != UIO_USERSPACE || uio->uio_td == curthread, ("ncl_write proc")); if (vp->v_type != VREG) return (EIO); mtx_lock(&np->n_mtx); if (np->n_flag & NWRITEERR) { np->n_flag &= ~NWRITEERR; mtx_unlock(&np->n_mtx); return (np->n_error); } else mtx_unlock(&np->n_mtx); mtx_lock(&nmp->nm_mtx); if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { mtx_unlock(&nmp->nm_mtx); (void)ncl_fsinfo(nmp, vp, cred, td); mtx_lock(&nmp->nm_mtx); } if (nmp->nm_wsize == 0) (void) newnfs_iosize(nmp); mtx_unlock(&nmp->nm_mtx); /* * Synchronously flush pending buffers if we are in synchronous * mode or if we are appending. */ if (ioflag & (IO_APPEND | IO_SYNC)) { mtx_lock(&np->n_mtx); if (np->n_flag & NMODIFIED) { mtx_unlock(&np->n_mtx); #ifdef notyet /* Needs matching nonblock semantics elsewhere, too. */ /* * Require non-blocking, synchronous writes to * dirty files to inform the program it needs * to fsync(2) explicitly. */ if (ioflag & IO_NDELAY) return (EAGAIN); #endif np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); error = ncl_vinvalbuf(vp, V_SAVE | ((ioflag & IO_VMIO) != 0 ? V_VMIO : 0), td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) error = EBADF; if (error != 0) return (error); } else mtx_unlock(&np->n_mtx); } orig_resid = uio->uio_resid; mtx_lock(&np->n_mtx); orig_size = np->n_size; mtx_unlock(&np->n_mtx); /* * If IO_APPEND then load uio_offset. We restart here if we cannot * get the append lock. */ if (ioflag & IO_APPEND) { np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); error = VOP_GETATTR(vp, &vattr, cred); if (error) return (error); mtx_lock(&np->n_mtx); uio->uio_offset = np->n_size; mtx_unlock(&np->n_mtx); } if (uio->uio_offset < 0) return (EINVAL); tmp_off = uio->uio_offset + uio->uio_resid; if (tmp_off > nmp->nm_maxfilesize || tmp_off < uio->uio_offset) return (EFBIG); if (uio->uio_resid == 0) return (0); if (newnfs_directio_enable && (ioflag & IO_DIRECT) && vp->v_type == VREG) return nfs_directio_write(vp, uio, cred, ioflag); /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, i don't think it matters */ if (vn_rlimit_fsize(vp, uio, td)) return (EFBIG); biosize = vp->v_bufobj.bo_bsize; /* * Find all of this file's B_NEEDCOMMIT buffers. If our writes * would exceed the local maximum per-file write commit size when * combined with those, we must decide whether to flush, * go synchronous, or return error. We don't bother checking * IO_UNIT -- we just make all writes atomic anyway, as there's * no point optimizing for something that really won't ever happen. */ wouldcommit = 0; if (!(ioflag & IO_SYNC)) { int nflag; mtx_lock(&np->n_mtx); nflag = np->n_flag; mtx_unlock(&np->n_mtx); if (nflag & NMODIFIED) { BO_LOCK(&vp->v_bufobj); if (vp->v_bufobj.bo_dirty.bv_cnt != 0) { TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) { if (bp->b_flags & B_NEEDCOMMIT) wouldcommit += bp->b_bcount; } } BO_UNLOCK(&vp->v_bufobj); } } do { if (!(ioflag & IO_SYNC)) { wouldcommit += biosize; if (wouldcommit > nmp->nm_wcommitsize) { np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); error = ncl_vinvalbuf(vp, V_SAVE | ((ioflag & IO_VMIO) != 0 ? V_VMIO : 0), td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) error = EBADF; if (error != 0) return (error); wouldcommit = biosize; } } NFSINCRGLOBAL(nfsstatsv1.biocache_writes); lbn = uio->uio_offset / biosize; on = uio->uio_offset - (lbn * biosize); n = MIN((unsigned)(biosize - on), uio->uio_resid); again: /* * Handle direct append and file extension cases, calculate * unaligned buffer size. */ mtx_lock(&np->n_mtx); if ((np->n_flag & NHASBEENLOCKED) == 0 && (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0) noncontig_write = 1; else noncontig_write = 0; if ((uio->uio_offset == np->n_size || (noncontig_write != 0 && lbn == (np->n_size / biosize) && uio->uio_offset + n > np->n_size)) && n) { mtx_unlock(&np->n_mtx); /* * Get the buffer (in its pre-append state to maintain * B_CACHE if it was previously set). Resize the * nfsnode after we have locked the buffer to prevent * readers from reading garbage. */ obcount = np->n_size - (lbn * biosize); bp = nfs_getcacheblk(vp, lbn, obcount, td); if (bp != NULL) { long save; mtx_lock(&np->n_mtx); np->n_size = uio->uio_offset + n; np->n_flag |= NMODIFIED; vnode_pager_setsize(vp, np->n_size); mtx_unlock(&np->n_mtx); save = bp->b_flags & B_CACHE; bcount = on + n; allocbuf(bp, bcount); bp->b_flags |= save; if (noncontig_write != 0 && on > obcount) vfs_bio_bzero_buf(bp, obcount, on - obcount); } } else { /* * Obtain the locked cache block first, and then * adjust the file's size as appropriate. */ bcount = on + n; if ((off_t)lbn * biosize + bcount < np->n_size) { if ((off_t)(lbn + 1) * biosize < np->n_size) bcount = biosize; else bcount = np->n_size - (off_t)lbn * biosize; } mtx_unlock(&np->n_mtx); bp = nfs_getcacheblk(vp, lbn, bcount, td); mtx_lock(&np->n_mtx); if (uio->uio_offset + n > np->n_size) { np->n_size = uio->uio_offset + n; np->n_flag |= NMODIFIED; vnode_pager_setsize(vp, np->n_size); } mtx_unlock(&np->n_mtx); } if (!bp) { error = newnfs_sigintr(nmp, td); if (!error) error = EINTR; break; } /* * Issue a READ if B_CACHE is not set. In special-append * mode, B_CACHE is based on the buffer prior to the write * op and is typically set, avoiding the read. If a read * is required in special append mode, the server will * probably send us a short-read since we extended the file * on our end, resulting in b_resid == 0 and, thusly, * B_CACHE getting set. * * We can also avoid issuing the read if the write covers * the entire buffer. We have to make sure the buffer state * is reasonable in this case since we will not be initiating * I/O. See the comments in kern/vfs_bio.c's getblk() for * more information. * * B_CACHE may also be set due to the buffer being cached * normally. */ bp_cached = 1; if (on == 0 && n == bcount) { if ((bp->b_flags & B_CACHE) == 0) bp_cached = 0; bp->b_flags |= B_CACHE; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; } if ((bp->b_flags & B_CACHE) == 0) { bp->b_iocmd = BIO_READ; vfs_busy_pages(bp, 0); error = ncl_doio(vp, bp, cred, td, 0); if (error) { brelse(bp); break; } } if (bp->b_wcred == NOCRED) bp->b_wcred = crhold(cred); mtx_lock(&np->n_mtx); np->n_flag |= NMODIFIED; mtx_unlock(&np->n_mtx); /* * If dirtyend exceeds file size, chop it down. This should * not normally occur but there is an append race where it * might occur XXX, so we log it. * * If the chopping creates a reverse-indexed or degenerate * situation with dirtyoff/end, we 0 both of them. */ if (bp->b_dirtyend > bcount) { printf("NFS append race @%lx:%d\n", (long)bp->b_blkno * DEV_BSIZE, bp->b_dirtyend - bcount); bp->b_dirtyend = bcount; } if (bp->b_dirtyoff >= bp->b_dirtyend) bp->b_dirtyoff = bp->b_dirtyend = 0; /* * If the new write will leave a contiguous dirty * area, just update the b_dirtyoff and b_dirtyend, * otherwise force a write rpc of the old dirty area. * * If there has been a file lock applied to this file * or vfs.nfs.old_noncontig_writing is set, do the following: * While it is possible to merge discontiguous writes due to * our having a B_CACHE buffer ( and thus valid read data * for the hole), we don't because it could lead to * significant cache coherency problems with multiple clients, * especially if locking is implemented later on. * * If vfs.nfs.old_noncontig_writing is not set and there has * not been file locking done on this file: * Relax coherency a bit for the sake of performance and * expand the current dirty region to contain the new * write even if it means we mark some non-dirty data as * dirty. */ if (noncontig_write == 0 && bp->b_dirtyend > 0 && (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) { if (bwrite(bp) == EINTR) { error = EINTR; break; } goto again; } local_resid = uio->uio_resid; error = vn_io_fault_uiomove((char *)bp->b_data + on, n, uio); if (error != 0 && !bp_cached) { /* * This block has no other content then what * possibly was written by the faulty uiomove. * Release it, forgetting the data pages, to * prevent the leak of uninitialized data to * usermode. */ bp->b_ioflags |= BIO_ERROR; brelse(bp); uio->uio_offset -= local_resid - uio->uio_resid; uio->uio_resid = local_resid; break; } /* * Since this block is being modified, it must be written * again and not just committed. Since write clustering does * not work for the stage 1 data write, only the stage 2 * commit rpc, we have to clear B_CLUSTEROK as well. */ bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); /* * Get the partial update on the progress made from * uiomove, if an error occurred. */ if (error != 0) n = local_resid - uio->uio_resid; /* * Only update dirtyoff/dirtyend if not a degenerate * condition. */ if (n > 0) { if (bp->b_dirtyend > 0) { bp->b_dirtyoff = min(on, bp->b_dirtyoff); bp->b_dirtyend = max((on + n), bp->b_dirtyend); } else { bp->b_dirtyoff = on; bp->b_dirtyend = on + n; } vfs_bio_set_valid(bp, on, n); } /* * If IO_SYNC do bwrite(). * * IO_INVAL appears to be unused. The idea appears to be * to turn off caching in this case. Very odd. XXX */ if ((ioflag & IO_SYNC)) { if (ioflag & IO_INVAL) bp->b_flags |= B_NOCACHE; error1 = bwrite(bp); if (error1 != 0) { if (error == 0) error = error1; break; } } else if ((n + on) == biosize || (ioflag & IO_ASYNC) != 0) { bp->b_flags |= B_ASYNC; (void) ncl_writebp(bp, 0, NULL); } else { bdwrite(bp); } if (error != 0) break; } while (uio->uio_resid > 0 && n > 0); if (error != 0) { if (ioflag & IO_UNIT) { VATTR_NULL(&vattr); vattr.va_size = orig_size; /* IO_SYNC is handled implicitely */ (void)VOP_SETATTR(vp, &vattr, cred); uio->uio_offset -= orig_resid - uio->uio_resid; uio->uio_resid = orig_resid; } } return (error); } /* * Get an nfs cache block. * * Allocate a new one if the block isn't currently in the cache * and return the block marked busy. If the calling process is * interrupted by a signal for an interruptible mount point, return * NULL. * * The caller must carefully deal with the possible B_INVAL state of * the buffer. ncl_doio() clears B_INVAL (and ncl_asyncio() clears it * indirectly), so synchronous reads can be issued without worrying about * the B_INVAL state. We have to be a little more careful when dealing * with writes (see comments in nfs_write()) when extending a file past * its EOF. */ static struct buf * nfs_getcacheblk(struct vnode *vp, daddr_t bn, int size, struct thread *td) { struct buf *bp; struct mount *mp; struct nfsmount *nmp; mp = vp->v_mount; nmp = VFSTONFS(mp); if (nmp->nm_flag & NFSMNT_INT) { sigset_t oldset; newnfs_set_sigmask(td, &oldset); bp = getblk(vp, bn, size, PCATCH, 0, 0); newnfs_restore_sigmask(td, &oldset); while (bp == NULL) { if (newnfs_sigintr(nmp, td)) return (NULL); bp = getblk(vp, bn, size, 0, 2 * hz, 0); } } else { bp = getblk(vp, bn, size, 0, 0, 0); } if (vp->v_type == VREG) bp->b_blkno = bn * (vp->v_bufobj.bo_bsize / DEV_BSIZE); return (bp); } /* * Flush and invalidate all dirty buffers. If another process is already * doing the flush, just wait for completion. */ int ncl_vinvalbuf(struct vnode *vp, int flags, struct thread *td, int intrflg) { struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, slpflag, slptimeo; int old_lock = 0; ASSERT_VOP_LOCKED(vp, "ncl_vinvalbuf"); if ((nmp->nm_flag & NFSMNT_INT) == 0) intrflg = 0; - if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)) + if (NFSCL_FORCEDISM(nmp->nm_mountp)) intrflg = 1; if (intrflg) { slpflag = PCATCH; slptimeo = 2 * hz; } else { slpflag = 0; slptimeo = 0; } old_lock = ncl_upgrade_vnlock(vp); if (vp->v_iflag & VI_DOOMED) { /* * Since vgonel() uses the generic vinvalbuf() to flush * dirty buffers and it does not call this function, it * is safe to just return OK when VI_DOOMED is set. */ ncl_downgrade_vnlock(vp, old_lock); return (0); } /* * Now, flush as required. */ if ((flags & (V_SAVE | V_VMIO)) == V_SAVE && vp->v_bufobj.bo_object != NULL) { VM_OBJECT_WLOCK(vp->v_bufobj.bo_object); vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_bufobj.bo_object); /* * If the page clean was interrupted, fail the invalidation. * Not doing so, we run the risk of losing dirty pages in the * vinvalbuf() call below. */ if (intrflg && (error = newnfs_sigintr(nmp, td))) goto out; } error = vinvalbuf(vp, flags, slpflag, 0); while (error) { if (intrflg && (error = newnfs_sigintr(nmp, td))) goto out; error = vinvalbuf(vp, flags, 0, slptimeo); } if (NFSHASPNFS(nmp)) { nfscl_layoutcommit(vp, td); /* * Invalidate the attribute cache, since writes to a DS * won't update the size attribute. */ mtx_lock(&np->n_mtx); np->n_attrstamp = 0; } else mtx_lock(&np->n_mtx); if (np->n_directio_asyncwr == 0) np->n_flag &= ~NMODIFIED; mtx_unlock(&np->n_mtx); out: ncl_downgrade_vnlock(vp, old_lock); return error; } /* * Initiate asynchronous I/O. Return an error if no nfsiods are available. * This is mainly to avoid queueing async I/O requests when the nfsiods * are all hung on a dead server. * * Note: ncl_asyncio() does not clear (BIO_ERROR|B_INVAL) but when the bp * is eventually dequeued by the async daemon, ncl_doio() *will*. */ int ncl_asyncio(struct nfsmount *nmp, struct buf *bp, struct ucred *cred, struct thread *td) { int iod; int gotiod; int slpflag = 0; int slptimeo = 0; int error, error2; /* * Commits are usually short and sweet so lets save some cpu and * leave the async daemons for more important rpc's (such as reads * and writes). * * Readdirplus RPCs do vget()s to acquire the vnodes for entries * in the directory in order to update attributes. This can deadlock * with another thread that is waiting for async I/O to be done by * an nfsiod thread while holding a lock on one of these vnodes. * To avoid this deadlock, don't allow the async nfsiod threads to * perform Readdirplus RPCs. */ mtx_lock(&ncl_iod_mutex); if ((bp->b_iocmd == BIO_WRITE && (bp->b_flags & B_NEEDCOMMIT) && (nmp->nm_bufqiods > ncl_numasync / 2)) || (bp->b_vp->v_type == VDIR && (nmp->nm_flag & NFSMNT_RDIRPLUS))) { mtx_unlock(&ncl_iod_mutex); return(EIO); } again: if (nmp->nm_flag & NFSMNT_INT) slpflag = PCATCH; gotiod = FALSE; /* * Find a free iod to process this request. */ for (iod = 0; iod < ncl_numasync; iod++) if (ncl_iodwant[iod] == NFSIOD_AVAILABLE) { gotiod = TRUE; break; } /* * Try to create one if none are free. */ if (!gotiod) ncl_nfsiodnew(); else { /* * Found one, so wake it up and tell it which * mount to process. */ NFS_DPF(ASYNCIO, ("ncl_asyncio: waking iod %d for mount %p\n", iod, nmp)); ncl_iodwant[iod] = NFSIOD_NOT_AVAILABLE; ncl_iodmount[iod] = nmp; nmp->nm_bufqiods++; wakeup(&ncl_iodwant[iod]); } /* * If none are free, we may already have an iod working on this mount * point. If so, it will process our request. */ if (!gotiod) { if (nmp->nm_bufqiods > 0) { NFS_DPF(ASYNCIO, ("ncl_asyncio: %d iods are already processing mount %p\n", nmp->nm_bufqiods, nmp)); gotiod = TRUE; } } /* * If we have an iod which can process the request, then queue * the buffer. */ if (gotiod) { /* * Ensure that the queue never grows too large. We still want * to asynchronize so we block rather then return EIO. */ while (nmp->nm_bufqlen >= 2*ncl_numasync) { NFS_DPF(ASYNCIO, ("ncl_asyncio: waiting for mount %p queue to drain\n", nmp)); nmp->nm_bufqwant = TRUE; error = newnfs_msleep(td, &nmp->nm_bufq, &ncl_iod_mutex, slpflag | PRIBIO, "nfsaio", slptimeo); if (error) { error2 = newnfs_sigintr(nmp, td); if (error2) { mtx_unlock(&ncl_iod_mutex); return (error2); } if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } } /* * We might have lost our iod while sleeping, * so check and loop if necessary. */ goto again; } /* We might have lost our nfsiod */ if (nmp->nm_bufqiods == 0) { NFS_DPF(ASYNCIO, ("ncl_asyncio: no iods after mount %p queue was drained, looping\n", nmp)); goto again; } if (bp->b_iocmd == BIO_READ) { if (bp->b_rcred == NOCRED && cred != NOCRED) bp->b_rcred = crhold(cred); } else { if (bp->b_wcred == NOCRED && cred != NOCRED) bp->b_wcred = crhold(cred); } if (bp->b_flags & B_REMFREE) bremfreef(bp); BUF_KERNPROC(bp); TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist); nmp->nm_bufqlen++; if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) { mtx_lock(&(VTONFS(bp->b_vp))->n_mtx); VTONFS(bp->b_vp)->n_flag |= NMODIFIED; VTONFS(bp->b_vp)->n_directio_asyncwr++; mtx_unlock(&(VTONFS(bp->b_vp))->n_mtx); } mtx_unlock(&ncl_iod_mutex); return (0); } mtx_unlock(&ncl_iod_mutex); /* * All the iods are busy on other mounts, so return EIO to * force the caller to process the i/o synchronously. */ NFS_DPF(ASYNCIO, ("ncl_asyncio: no iods available, i/o is synchronous\n")); return (EIO); } void ncl_doio_directwrite(struct buf *bp) { int iomode, must_commit; struct uio *uiop = (struct uio *)bp->b_caller1; char *iov_base = uiop->uio_iov->iov_base; iomode = NFSWRITE_FILESYNC; uiop->uio_td = NULL; /* NULL since we're in nfsiod */ ncl_writerpc(bp->b_vp, uiop, bp->b_wcred, &iomode, &must_commit, 0); KASSERT((must_commit == 0), ("ncl_doio_directwrite: Did not commit write")); free(iov_base, M_NFSDIRECTIO); free(uiop->uio_iov, M_NFSDIRECTIO); free(uiop, M_NFSDIRECTIO); if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) { struct nfsnode *np = VTONFS(bp->b_vp); mtx_lock(&np->n_mtx); if (NFSHASPNFS(VFSTONFS(vnode_mount(bp->b_vp)))) { /* * Invalidate the attribute cache, since writes to a DS * won't update the size attribute. */ np->n_attrstamp = 0; } np->n_directio_asyncwr--; if (np->n_directio_asyncwr == 0) { np->n_flag &= ~NMODIFIED; if ((np->n_flag & NFSYNCWAIT)) { np->n_flag &= ~NFSYNCWAIT; wakeup((caddr_t)&np->n_directio_asyncwr); } } mtx_unlock(&np->n_mtx); } bp->b_vp = NULL; relpbuf(bp, &ncl_pbuf_freecnt); } /* * Do an I/O operation to/from a cache block. This may be called * synchronously or from an nfsiod. */ int ncl_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td, int called_from_strategy) { struct uio *uiop; struct nfsnode *np; struct nfsmount *nmp; int error = 0, iomode, must_commit = 0; struct uio uio; struct iovec io; struct proc *p = td ? td->td_proc : NULL; uint8_t iocmd; np = VTONFS(vp); nmp = VFSTONFS(vp->v_mount); uiop = &uio; uiop->uio_iov = &io; uiop->uio_iovcnt = 1; uiop->uio_segflg = UIO_SYSSPACE; uiop->uio_td = td; /* * clear BIO_ERROR and B_INVAL state prior to initiating the I/O. We * do this here so we do not have to do it in all the code that * calls us. */ bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; KASSERT(!(bp->b_flags & B_DONE), ("ncl_doio: bp %p already marked done", bp)); iocmd = bp->b_iocmd; if (iocmd == BIO_READ) { io.iov_len = uiop->uio_resid = bp->b_bcount; io.iov_base = bp->b_data; uiop->uio_rw = UIO_READ; switch (vp->v_type) { case VREG: uiop->uio_offset = ((off_t)bp->b_blkno) * DEV_BSIZE; NFSINCRGLOBAL(nfsstatsv1.read_bios); error = ncl_readrpc(vp, uiop, cr); if (!error) { if (uiop->uio_resid) { /* * If we had a short read with no error, we must have * hit a file hole. We should zero-fill the remainder. * This can also occur if the server hits the file EOF. * * Holes used to be able to occur due to pending * writes, but that is not possible any longer. */ int nread = bp->b_bcount - uiop->uio_resid; ssize_t left = uiop->uio_resid; if (left > 0) bzero((char *)bp->b_data + nread, left); uiop->uio_resid = 0; } } /* ASSERT_VOP_LOCKED(vp, "ncl_doio"); */ if (p && (vp->v_vflag & VV_TEXT)) { mtx_lock(&np->n_mtx); if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &np->n_vattr.na_mtime)) { mtx_unlock(&np->n_mtx); PROC_LOCK(p); killproc(p, "text file modification"); PROC_UNLOCK(p); } else mtx_unlock(&np->n_mtx); } break; case VLNK: uiop->uio_offset = (off_t)0; NFSINCRGLOBAL(nfsstatsv1.readlink_bios); error = ncl_readlinkrpc(vp, uiop, cr); break; case VDIR: NFSINCRGLOBAL(nfsstatsv1.readdir_bios); uiop->uio_offset = ((u_quad_t)bp->b_lblkno) * NFS_DIRBLKSIZ; if ((nmp->nm_flag & NFSMNT_RDIRPLUS) != 0) { error = ncl_readdirplusrpc(vp, uiop, cr, td); if (error == NFSERR_NOTSUPP) nmp->nm_flag &= ~NFSMNT_RDIRPLUS; } if ((nmp->nm_flag & NFSMNT_RDIRPLUS) == 0) error = ncl_readdirrpc(vp, uiop, cr, td); /* * end-of-directory sets B_INVAL but does not generate an * error. */ if (error == 0 && uiop->uio_resid == bp->b_bcount) bp->b_flags |= B_INVAL; break; default: printf("ncl_doio: type %x unexpected\n", vp->v_type); break; } if (error) { bp->b_ioflags |= BIO_ERROR; bp->b_error = error; } } else { /* * If we only need to commit, try to commit */ if (bp->b_flags & B_NEEDCOMMIT) { int retv; off_t off; off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; retv = ncl_commit(vp, off, bp->b_dirtyend-bp->b_dirtyoff, bp->b_wcred, td); if (retv == 0) { bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); bp->b_resid = 0; bufdone(bp); return (0); } if (retv == NFSERR_STALEWRITEVERF) { ncl_clearcommit(vp->v_mount); } } /* * Setup for actual write */ mtx_lock(&np->n_mtx); if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size) bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE; mtx_unlock(&np->n_mtx); if (bp->b_dirtyend > bp->b_dirtyoff) { io.iov_len = uiop->uio_resid = bp->b_dirtyend - bp->b_dirtyoff; uiop->uio_offset = (off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyoff; io.iov_base = (char *)bp->b_data + bp->b_dirtyoff; uiop->uio_rw = UIO_WRITE; NFSINCRGLOBAL(nfsstatsv1.write_bios); if ((bp->b_flags & (B_ASYNC | B_NEEDCOMMIT | B_NOCACHE | B_CLUSTER)) == B_ASYNC) iomode = NFSWRITE_UNSTABLE; else iomode = NFSWRITE_FILESYNC; error = ncl_writerpc(vp, uiop, cr, &iomode, &must_commit, called_from_strategy); /* * When setting B_NEEDCOMMIT also set B_CLUSTEROK to try * to cluster the buffers needing commit. This will allow * the system to submit a single commit rpc for the whole * cluster. We can do this even if the buffer is not 100% * dirty (relative to the NFS blocksize), so we optimize the * append-to-file-case. * * (when clearing B_NEEDCOMMIT, B_CLUSTEROK must also be * cleared because write clustering only works for commit * rpc's, not for the data portion of the write). */ if (!error && iomode == NFSWRITE_UNSTABLE) { bp->b_flags |= B_NEEDCOMMIT; if (bp->b_dirtyoff == 0 && bp->b_dirtyend == bp->b_bcount) bp->b_flags |= B_CLUSTEROK; } else { bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); } /* * For an interrupted write, the buffer is still valid * and the write hasn't been pushed to the server yet, * so we can't set BIO_ERROR and report the interruption * by setting B_EINTR. For the B_ASYNC case, B_EINTR * is not relevant, so the rpc attempt is essentially * a noop. For the case of a V3 write rpc not being * committed to stable storage, the block is still * dirty and requires either a commit rpc or another * write rpc with iomode == NFSV3WRITE_FILESYNC before * the block is reused. This is indicated by setting * the B_DELWRI and B_NEEDCOMMIT flags. * * EIO is returned by ncl_writerpc() to indicate a recoverable * write error and is handled as above, except that * B_EINTR isn't set. One cause of this is a stale stateid * error for the RPC that indicates recovery is required, * when called with called_from_strategy != 0. * * If the buffer is marked B_PAGING, it does not reside on * the vp's paging queues so we cannot call bdirty(). The * bp in this case is not an NFS cache block so we should * be safe. XXX * * The logic below breaks up errors into recoverable and * unrecoverable. For the former, we clear B_INVAL|B_NOCACHE * and keep the buffer around for potential write retries. * For the latter (eg ESTALE), we toss the buffer away (B_INVAL) * and save the error in the nfsnode. This is less than ideal * but necessary. Keeping such buffers around could potentially * cause buffer exhaustion eventually (they can never be written * out, so will get constantly be re-dirtied). It also causes * all sorts of vfs panics. For non-recoverable write errors, * also invalidate the attrcache, so we'll be forced to go over * the wire for this object, returning an error to user on next * call (most of the time). */ if (error == EINTR || error == EIO || error == ETIMEDOUT || (!error && (bp->b_flags & B_NEEDCOMMIT))) { bp->b_flags &= ~(B_INVAL|B_NOCACHE); if ((bp->b_flags & B_PAGING) == 0) { bdirty(bp); bp->b_flags &= ~B_DONE; } if ((error == EINTR || error == ETIMEDOUT) && (bp->b_flags & B_ASYNC) == 0) bp->b_flags |= B_EINTR; } else { if (error) { bp->b_ioflags |= BIO_ERROR; bp->b_flags |= B_INVAL; bp->b_error = np->n_error = error; mtx_lock(&np->n_mtx); np->n_flag |= NWRITEERR; np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); mtx_unlock(&np->n_mtx); } bp->b_dirtyoff = bp->b_dirtyend = 0; } } else { bp->b_resid = 0; bufdone(bp); return (0); } } bp->b_resid = uiop->uio_resid; if (must_commit) ncl_clearcommit(vp->v_mount); bufdone(bp); return (error); } /* * Used to aid in handling ftruncate() operations on the NFS client side. * Truncation creates a number of special problems for NFS. We have to * throw away VM pages and buffer cache buffers that are beyond EOF, and * we have to properly handle VM pages or (potentially dirty) buffers * that straddle the truncation point. */ int ncl_meta_setsize(struct vnode *vp, struct ucred *cred, struct thread *td, u_quad_t nsize) { struct nfsnode *np = VTONFS(vp); u_quad_t tsize; int biosize = vp->v_bufobj.bo_bsize; int error = 0; mtx_lock(&np->n_mtx); tsize = np->n_size; np->n_size = nsize; mtx_unlock(&np->n_mtx); if (nsize < tsize) { struct buf *bp; daddr_t lbn; int bufsize; /* * vtruncbuf() doesn't get the buffer overlapping the * truncation point. We may have a B_DELWRI and/or B_CACHE * buffer that now needs to be truncated. */ error = vtruncbuf(vp, cred, nsize, biosize); lbn = nsize / biosize; bufsize = nsize - (lbn * biosize); bp = nfs_getcacheblk(vp, lbn, bufsize, td); if (!bp) return EINTR; if (bp->b_dirtyoff > bp->b_bcount) bp->b_dirtyoff = bp->b_bcount; if (bp->b_dirtyend > bp->b_bcount) bp->b_dirtyend = bp->b_bcount; bp->b_flags |= B_RELBUF; /* don't leave garbage around */ brelse(bp); } else { vnode_pager_setsize(vp, nsize); } return(error); } Index: head/sys/fs/nfsclient/nfs_clport.c =================================================================== --- head/sys/fs/nfsclient/nfs_clport.c (revision 321627) +++ head/sys/fs/nfsclient/nfs_clport.c (revision 321628) @@ -1,1438 +1,1438 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include /* * generally, I don't like #includes inside .h files, but it seems to * be the easiest way to handle the port. */ #include #include #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS dtrace_nfsclient_attrcache_flush_probe_func_t dtrace_nfscl_attrcache_flush_done_probe; uint32_t nfscl_attrcache_flush_done_id; dtrace_nfsclient_attrcache_get_hit_probe_func_t dtrace_nfscl_attrcache_get_hit_probe; uint32_t nfscl_attrcache_get_hit_id; dtrace_nfsclient_attrcache_get_miss_probe_func_t dtrace_nfscl_attrcache_get_miss_probe; uint32_t nfscl_attrcache_get_miss_id; dtrace_nfsclient_attrcache_load_probe_func_t dtrace_nfscl_attrcache_load_done_probe; uint32_t nfscl_attrcache_load_done_id; #endif /* !KDTRACE_HOOKS */ extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1; extern struct vop_vector newnfs_vnodeops; extern struct vop_vector newnfs_fifoops; extern uma_zone_t newnfsnode_zone; extern struct buf_ops buf_ops_newnfs; extern int ncl_pbuf_freecnt; extern short nfsv4_cbport; extern int nfscl_enablecallb; extern int nfs_numnfscbd; extern int nfscl_inited; struct mtx ncl_iod_mutex; NFSDLOCKMUTEX; extern void (*ncl_call_invalcaches)(struct vnode *); SYSCTL_DECL(_vfs_nfs); static int ncl_fileid_maxwarnings = 10; SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN, &ncl_fileid_maxwarnings, 0, "Limit fileid corruption warnings; 0 is off; -1 is unlimited"); static volatile int ncl_fileid_nwarnings; static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *, struct nfsvattr *); /* * Comparison function for vfs_hash functions. */ int newnfs_vncmpf(struct vnode *vp, void *arg) { struct nfsfh *nfhp = (struct nfsfh *)arg; struct nfsnode *np = VTONFS(vp); if (np->n_fhp->nfh_len != nfhp->nfh_len || NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len)) return (1); return (0); } /* * Look up a vnode/nfsnode by file handle. * Callers must check for mount points!! * In all cases, a pointer to a * nfsnode structure is returned. * This variant takes a "struct nfsfh *" as second argument and uses * that structure up, either by hanging off the nfsnode or FREEing it. */ int nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp, struct componentname *cnp, struct thread *td, struct nfsnode **npp, void *stuff, int lkflags) { struct nfsnode *np, *dnp; struct vnode *vp, *nvp; struct nfsv4node *newd, *oldd; int error; u_int hash; struct nfsmount *nmp; nmp = VFSTONFS(mntp); dnp = VTONFS(dvp); *npp = NULL; hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); error = vfs_hash_get(mntp, hash, lkflags, td, &nvp, newnfs_vncmpf, nfhp); if (error == 0 && nvp != NULL) { /* * I believe there is a slight chance that vgonel() could * get called on this vnode between when NFSVOPLOCK() drops * the VI_LOCK() and vget() acquires it again, so that it * hasn't yet had v_usecount incremented. If this were to * happen, the VI_DOOMED flag would be set, so check for * that here. Since we now have the v_usecount incremented, * we should be ok until we vrele() it, if the VI_DOOMED * flag isn't set now. */ VI_LOCK(nvp); if ((nvp->v_iflag & VI_DOOMED)) { VI_UNLOCK(nvp); vrele(nvp); error = ENOENT; } else { VI_UNLOCK(nvp); } } if (error) { FREE((caddr_t)nfhp, M_NFSFH); return (error); } if (nvp != NULL) { np = VTONFS(nvp); /* * For NFSv4, check to see if it is the same name and * replace the name, if it is different. */ oldd = newd = NULL; if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL && nvp->v_type == VREG && (np->n_v4->n4_namelen != cnp->cn_namelen || NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen) || dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len))) { MALLOC(newd, struct nfsv4node *, sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); NFSLOCKNODE(np); if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG && (np->n_v4->n4_namelen != cnp->cn_namelen || NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen) || dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len))) { oldd = np->n_v4; np->n_v4 = newd; newd = NULL; np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; np->n_v4->n4_namelen = cnp->cn_namelen; NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len); NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen); } NFSUNLOCKNODE(np); } if (newd != NULL) FREE((caddr_t)newd, M_NFSV4NODE); if (oldd != NULL) FREE((caddr_t)oldd, M_NFSV4NODE); *npp = np; FREE((caddr_t)nfhp, M_NFSFH); return (0); } np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO); error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp); if (error) { uma_zfree(newnfsnode_zone, np); FREE((caddr_t)nfhp, M_NFSFH); return (error); } vp = nvp; KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0")); vp->v_bufobj.bo_ops = &buf_ops_newnfs; vp->v_data = np; np->n_vnode = vp; /* * Initialize the mutex even if the vnode is going to be a loser. * This simplifies the logic in reclaim, which can then unconditionally * destroy the mutex (in the case of the loser, or if hash_insert * happened to return an error no special casing is needed). */ mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK); /* * Are we getting the root? If so, make sure the vnode flags * are correct */ if ((nfhp->nfh_len == nmp->nm_fhsize) && !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) { if (vp->v_type == VNON) vp->v_type = VDIR; vp->v_vflag |= VV_ROOT; } np->n_fhp = nfhp; /* * For NFSv4, we have to attach the directory file handle and * file name, so that Open Ops can be done later. */ if (nmp->nm_flag & NFSMNT_NFSV4) { MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; np->n_v4->n4_namelen = cnp->cn_namelen; NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len); NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4), cnp->cn_namelen); } else { np->n_v4 = NULL; } /* * NFS supports recursive and shared locking. */ lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL); VN_LOCK_AREC(vp); VN_LOCK_ASHARE(vp); error = insmntque(vp, mntp); if (error != 0) { *npp = NULL; mtx_destroy(&np->n_mtx); FREE((caddr_t)nfhp, M_NFSFH); if (np->n_v4 != NULL) FREE((caddr_t)np->n_v4, M_NFSV4NODE); uma_zfree(newnfsnode_zone, np); return (error); } error = vfs_hash_insert(vp, hash, lkflags, td, &nvp, newnfs_vncmpf, nfhp); if (error) return (error); if (nvp != NULL) { *npp = VTONFS(nvp); /* vfs_hash_insert() vput()'s the losing vnode */ return (0); } *npp = np; return (0); } /* * Another variant of nfs_nget(). This one is only used by reopen. It * takes almost the same args as nfs_nget(), but only succeeds if an entry * exists in the cache. (Since files should already be "open" with a * vnode ref cnt on the node when reopen calls this, it should always * succeed.) * Also, don't get a vnode lock, since it may already be locked by some * other process that is handling it. This is ok, since all other threads * on the client are blocked by the nfsc_lock being exclusively held by the * caller of this function. */ int nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize, struct thread *td, struct nfsnode **npp) { struct vnode *nvp; u_int hash; struct nfsfh *nfhp; int error; *npp = NULL; /* For forced dismounts, just return error. */ - if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF)) + if (NFSCL_FORCEDISM(mntp)) return (EINTR); MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize, M_NFSFH, M_WAITOK); bcopy(fhp, &nfhp->nfh_fh[0], fhsize); nfhp->nfh_len = fhsize; hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT); /* * First, try to get the vnode locked, but don't block for the lock. */ error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp, newnfs_vncmpf, nfhp); if (error == 0 && nvp != NULL) { NFSVOPUNLOCK(nvp, 0); } else if (error == EBUSY) { /* * It is safe so long as a vflush() with * FORCECLOSE has not been done. Since the Renew thread is * stopped and the MNTK_UNMOUNTF flag is set before doing * a vflush() with FORCECLOSE, we should be ok here. */ - if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF)) + if (NFSCL_FORCEDISM(mntp)) error = EINTR; else { vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp); if (nvp == NULL) { error = ENOENT; } else if ((nvp->v_iflag & VI_DOOMED) != 0) { error = ENOENT; vrele(nvp); } else { error = 0; } } } FREE(nfhp, M_NFSFH); if (error) return (error); if (nvp != NULL) { *npp = VTONFS(nvp); return (0); } return (EINVAL); } static void nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap, struct nfsvattr *newnap) { int off; if (ncl_fileid_maxwarnings >= 0 && ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) return; off = 0; if (ncl_fileid_maxwarnings >= 0) { if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings) off = 1; } printf("newnfs: server '%s' error: fileid changed. " "fsid %jx:%jx: expected fileid %#jx, got %#jx. " "(BROKEN NFS SERVER OR MIDDLEWARE)\n", nmp->nm_com.nmcom_hostname, (uintmax_t)nmp->nm_fsid[0], (uintmax_t)nmp->nm_fsid[1], (uintmax_t)oldnap->na_fileid, (uintmax_t)newnap->na_fileid); if (off) printf("newnfs: Logged %d times about fileid corruption; " "going quiet to avoid spamming logs excessively. (Limit " "is: %d).\n", ncl_fileid_nwarnings, ncl_fileid_maxwarnings); } /* * Load the attribute cache (that lives in the nfsnode entry) with * the attributes of the second argument and * Iff vaper not NULL * copy the attributes to *vaper * Similar to nfs_loadattrcache(), except the attributes are passed in * instead of being parsed out of the mbuf list. */ int nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper, void *stuff, int writeattr, int dontshrink) { struct vnode *vp = *vpp; struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper; struct nfsnode *np; struct nfsmount *nmp; struct timespec mtime_save; u_quad_t nsize; int setnsize, error, force_fid_err; error = 0; setnsize = 0; nsize = 0; /* * If v_type == VNON it is a new node, so fill in the v_type, * n_mtime fields. Check to see if it represents a special * device, and if so, check for a possible alias. Once the * correct vnode has been obtained, fill in the rest of the * information. */ np = VTONFS(vp); NFSLOCKNODE(np); if (vp->v_type != nvap->va_type) { vp->v_type = nvap->va_type; if (vp->v_type == VFIFO) vp->v_op = &newnfs_fifoops; np->n_mtime = nvap->va_mtime; } nmp = VFSTONFS(vp->v_mount); vap = &np->n_vattr.na_vattr; mtime_save = vap->va_mtime; if (writeattr) { np->n_vattr.na_filerev = nap->na_filerev; np->n_vattr.na_size = nap->na_size; np->n_vattr.na_mtime = nap->na_mtime; np->n_vattr.na_ctime = nap->na_ctime; np->n_vattr.na_fsid = nap->na_fsid; np->n_vattr.na_mode = nap->na_mode; } else { force_fid_err = 0; KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning, force_fid_err); /* * BROKEN NFS SERVER OR MIDDLEWARE * * Certain NFS servers (certain old proprietary filers ca. * 2006) or broken middleboxes (e.g. WAN accelerator products) * will respond to GETATTR requests with results for a * different fileid. * * The WAN accelerator we've observed not only serves stale * cache results for a given file, it also occasionally serves * results for wholly different files. This causes surprising * problems; for example the cached size attribute of a file * may truncate down and then back up, resulting in zero * regions in file contents read by applications. We observed * this reliably with Clang and .c files during parallel build. * A pcap revealed packet fragmentation and GETATTR RPC * responses with wholly wrong fileids. */ if ((np->n_vattr.na_fileid != 0 && np->n_vattr.na_fileid != nap->na_fileid) || force_fid_err) { nfscl_warn_fileid(nmp, &np->n_vattr, nap); error = EIDRM; goto out; } NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr, sizeof (struct nfsvattr)); } /* * For NFSv4, if the node's fsid is not equal to the mount point's * fsid, return the low order 32bits of the node's fsid. This * allows getcwd(3) to work. There is a chance that the fsid might * be the same as a local fs, but since this is in an NFS mount * point, I don't think that will cause any problems? */ if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) && (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] || nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) { /* * va_fsid needs to be set to some value derived from * np->n_vattr.na_filesid that is not equal * vp->v_mount->mnt_stat.f_fsid[0], so that it changes * from the value used for the top level server volume * in the mounted subtree. */ vn_fsid(vp, vap); if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0]) vap->va_fsid = hash32_buf( np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0); } else vn_fsid(vp, vap); np->n_attrstamp = time_second; if (vap->va_size != np->n_size) { if (vap->va_type == VREG) { if (dontshrink && vap->va_size < np->n_size) { /* * We've been told not to shrink the file; * zero np->n_attrstamp to indicate that * the attributes are stale. */ vap->va_size = np->n_size; np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); vnode_pager_setsize(vp, np->n_size); } else if (np->n_flag & NMODIFIED) { /* * We've modified the file: Use the larger * of our size, and the server's size. */ if (vap->va_size < np->n_size) { vap->va_size = np->n_size; } else { np->n_size = vap->va_size; np->n_flag |= NSIZECHANGED; } vnode_pager_setsize(vp, np->n_size); } else if (vap->va_size < np->n_size) { /* * When shrinking the size, the call to * vnode_pager_setsize() cannot be done * with the mutex held, so delay it until * after the mtx_unlock call. */ nsize = np->n_size = vap->va_size; np->n_flag |= NSIZECHANGED; setnsize = 1; } else { np->n_size = vap->va_size; np->n_flag |= NSIZECHANGED; vnode_pager_setsize(vp, np->n_size); } } else { np->n_size = vap->va_size; } } /* * The following checks are added to prevent a race between (say) * a READDIR+ and a WRITE. * READDIR+, WRITE requests sent out. * READDIR+ resp, WRITE resp received on client. * However, the WRITE resp was handled before the READDIR+ resp * causing the post op attrs from the write to be loaded first * and the attrs from the READDIR+ to be loaded later. If this * happens, we have stale attrs loaded into the attrcache. * We detect this by for the mtime moving back. We invalidate the * attrcache when this happens. */ if (timespeccmp(&mtime_save, &vap->va_mtime, >)) { /* Size changed or mtime went backwards */ np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } if (vaper != NULL) { NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap)); if (np->n_flag & NCHG) { if (np->n_flag & NACC) vaper->va_atime = np->n_atim; if (np->n_flag & NUPD) vaper->va_mtime = np->n_mtim; } } out: #ifdef KDTRACE_HOOKS if (np->n_attrstamp != 0) KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error); #endif NFSUNLOCKNODE(np); if (setnsize) vnode_pager_setsize(vp, nsize); return (error); } /* * Fill in the client id name. For these bytes: * 1 - they must be unique * 2 - they should be persistent across client reboots * 1 is more critical than 2 * Use the mount point's unique id plus either the uuid or, if that * isn't set, random junk. */ void nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen) { int uuidlen; /* * First, put in the 64bit mount point identifier. */ if (idlen >= sizeof (u_int64_t)) { NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t)); cp += sizeof (u_int64_t); idlen -= sizeof (u_int64_t); } /* * If uuid is non-zero length, use it. */ uuidlen = strlen(uuid); if (uuidlen > 0 && idlen >= uuidlen) { NFSBCOPY(uuid, cp, uuidlen); cp += uuidlen; idlen -= uuidlen; } /* * This only normally happens if the uuid isn't set. */ while (idlen > 0) { *cp++ = (u_int8_t)(arc4random() % 256); idlen--; } } /* * Fill in a lock owner name. For now, pid + the process's creation time. */ void nfscl_filllockowner(void *id, u_int8_t *cp, int flags) { union { u_int32_t lval; u_int8_t cval[4]; } tl; struct proc *p; if (id == NULL) { /* Return the single open_owner of all 0 bytes. */ bzero(cp, NFSV4CL_LOCKNAMELEN); return; } if ((flags & F_POSIX) != 0) { p = (struct proc *)id; tl.lval = p->p_pid; *cp++ = tl.cval[0]; *cp++ = tl.cval[1]; *cp++ = tl.cval[2]; *cp++ = tl.cval[3]; tl.lval = p->p_stats->p_start.tv_sec; *cp++ = tl.cval[0]; *cp++ = tl.cval[1]; *cp++ = tl.cval[2]; *cp++ = tl.cval[3]; tl.lval = p->p_stats->p_start.tv_usec; *cp++ = tl.cval[0]; *cp++ = tl.cval[1]; *cp++ = tl.cval[2]; *cp = tl.cval[3]; } else if ((flags & F_FLOCK) != 0) { bcopy(&id, cp, sizeof(id)); bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id)); } else { printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n"); bzero(cp, NFSV4CL_LOCKNAMELEN); } } /* * Find the parent process for the thread passed in as an argument. * If none exists, return NULL, otherwise return a thread for the parent. * (Can be any of the threads, since it is only used for td->td_proc.) */ NFSPROC_T * nfscl_getparent(struct thread *td) { struct proc *p; struct thread *ptd; if (td == NULL) return (NULL); p = td->td_proc; if (p->p_pid == 0) return (NULL); p = p->p_pptr; if (p == NULL) return (NULL); ptd = TAILQ_FIRST(&p->p_threads); return (ptd); } /* * Start up the renew kernel thread. */ static void start_nfscl(void *arg) { struct nfsclclient *clp; struct thread *td; clp = (struct nfsclclient *)arg; td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads); nfscl_renewthread(clp, td); kproc_exit(0); } void nfscl_start_renewthread(struct nfsclclient *clp) { kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0, "nfscl"); } /* * Handle wcc_data. * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr * as the first Op after PutFH. * (For NFSv4, the postop attributes are after the Op, so they can't be * parsed here. A separate call to nfscl_postop_attr() is required.) */ int nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp, struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff) { u_int32_t *tl; struct nfsnode *np = VTONFS(vp); struct nfsvattr nfsva; int error = 0; if (wccflagp != NULL) *wccflagp = 0; if (nd->nd_flag & ND_NFSV3) { *flagp = 0; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); if (*tl == newnfs_true) { NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED); if (wccflagp != NULL) { mtx_lock(&np->n_mtx); *wccflagp = (np->n_mtime.tv_sec == fxdr_unsigned(u_int32_t, *(tl + 2)) && np->n_mtime.tv_nsec == fxdr_unsigned(u_int32_t, *(tl + 3))); mtx_unlock(&np->n_mtx); } } error = nfscl_postop_attr(nd, nap, flagp, stuff); if (wccflagp != NULL && *flagp == 0) *wccflagp = 0; } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) == (ND_NFSV4 | ND_V4WCCATTR)) { error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL); if (error) return (error); /* * Get rid of Op# and status for next op. */ NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*++tl) nd->nd_flag |= ND_NOMOREDATA; if (wccflagp != NULL && nfsva.na_vattr.va_mtime.tv_sec != 0) { mtx_lock(&np->n_mtx); *wccflagp = (np->n_mtime.tv_sec == nfsva.na_vattr.va_mtime.tv_sec && np->n_mtime.tv_nsec == nfsva.na_vattr.va_mtime.tv_sec); mtx_unlock(&np->n_mtx); } } nfsmout: return (error); } /* * Get postop attributes. */ int nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp, void *stuff) { u_int32_t *tl; int error = 0; *retp = 0; if (nd->nd_flag & ND_NOMOREDATA) return (error); if (nd->nd_flag & ND_NFSV3) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); *retp = fxdr_unsigned(int, *tl); } else if (nd->nd_flag & ND_NFSV4) { /* * For NFSv4, the postop attr are at the end, so no point * in looking if nd_repstat != 0. */ if (!nd->nd_repstat) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) /* should never happen since nd_repstat != 0 */ nd->nd_flag |= ND_NOMOREDATA; else *retp = 1; } } else if (!nd->nd_repstat) { /* For NFSv2, the attributes are here iff nd_repstat == 0 */ *retp = 1; } if (*retp) { error = nfsm_loadattr(nd, nap); if (error) *retp = 0; } nfsmout: return (error); } /* * Fill in the setable attributes. The full argument indicates whether * to fill in them all or just mode and time. */ void nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap, struct vnode *vp, int flags, u_int32_t rdev) { u_int32_t *tl; struct nfsv2_sattr *sp; nfsattrbit_t attrbits; switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) { case ND_NFSV2: NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR); if (vap->va_mode == (mode_t)VNOVAL) sp->sa_mode = newnfs_xdrneg1; else sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode); if (vap->va_uid == (uid_t)VNOVAL) sp->sa_uid = newnfs_xdrneg1; else sp->sa_uid = txdr_unsigned(vap->va_uid); if (vap->va_gid == (gid_t)VNOVAL) sp->sa_gid = newnfs_xdrneg1; else sp->sa_gid = txdr_unsigned(vap->va_gid); if (flags & NFSSATTR_SIZE0) sp->sa_size = 0; else if (flags & NFSSATTR_SIZENEG1) sp->sa_size = newnfs_xdrneg1; else if (flags & NFSSATTR_SIZERDEV) sp->sa_size = txdr_unsigned(rdev); else sp->sa_size = txdr_unsigned(vap->va_size); txdr_nfsv2time(&vap->va_atime, &sp->sa_atime); txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime); break; case ND_NFSV3: if (vap->va_mode != (mode_t)VNOVAL) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl = txdr_unsigned(vap->va_mode); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; } if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl = txdr_unsigned(vap->va_uid); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; } if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = newnfs_true; *tl = txdr_unsigned(vap->va_gid); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; } if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = newnfs_true; txdr_hyper(vap->va_size, tl); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; } if (vap->va_atime.tv_sec != VNOVAL) { if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); txdr_nfsv3time(&vap->va_atime, tl); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); } } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); } if (vap->va_mtime.tv_sec != VNOVAL) { if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT); txdr_nfsv3time(&vap->va_mtime, tl); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER); } } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE); } break; case ND_NFSV4: NFSZERO_ATTRBIT(&attrbits); if (vap->va_mode != (mode_t)VNOVAL) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE); if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER); if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP); if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE); if (vap->va_atime.tv_sec != VNOVAL) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET); if (vap->va_mtime.tv_sec != VNOVAL) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET); (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0, &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0); break; } } /* * nfscl_request() - mostly a wrapper for newnfs_request(). */ int nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p, struct ucred *cred, void *stuff) { int ret, vers; struct nfsmount *nmp; nmp = VFSTONFS(vp->v_mount); if (nd->nd_flag & ND_NFSV4) vers = NFS_VER4; else if (nd->nd_flag & ND_NFSV3) vers = NFS_VER3; else vers = NFS_VER2; ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, NFS_PROG, vers, NULL, 1, NULL, NULL); return (ret); } /* * fill in this bsden's variant of statfs using nfsstatfs. */ void nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs) { struct statfs *sbp = (struct statfs *)statfs; if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) { sbp->f_bsize = NFS_FABLKSIZE; sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE; sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE; /* * Although sf_abytes is uint64_t and f_bavail is int64_t, * the value after dividing by NFS_FABLKSIZE is small * enough that it will fit in 63bits, so it is ok to * assign it to f_bavail without fear that it will become * negative. */ sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE; sbp->f_files = sfp->sf_tfiles; /* Since f_ffree is int64_t, clip it to 63bits. */ if (sfp->sf_ffiles > INT64_MAX) sbp->f_ffree = INT64_MAX; else sbp->f_ffree = sfp->sf_ffiles; } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) { /* * The type casts to (int32_t) ensure that this code is * compatible with the old NFS client, in that it will * propagate bit31 to the high order bits. This may or may * not be correct for NFSv2, but since it is a legacy * environment, I'd rather retain backwards compatibility. */ sbp->f_bsize = (int32_t)sfp->sf_bsize; sbp->f_blocks = (int32_t)sfp->sf_blocks; sbp->f_bfree = (int32_t)sfp->sf_bfree; sbp->f_bavail = (int32_t)sfp->sf_bavail; sbp->f_files = 0; sbp->f_ffree = 0; } } /* * Use the fsinfo stuff to update the mount point. */ void nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp) { if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) && fsp->fs_wtpref >= NFS_FABLKSIZE) nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) & ~(NFS_FABLKSIZE - 1); if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) { nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1); if (nmp->nm_wsize == 0) nmp->nm_wsize = fsp->fs_wtmax; } if (nmp->nm_wsize < NFS_FABLKSIZE) nmp->nm_wsize = NFS_FABLKSIZE; if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) && fsp->fs_rtpref >= NFS_FABLKSIZE) nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) & ~(NFS_FABLKSIZE - 1); if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) { nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1); if (nmp->nm_rsize == 0) nmp->nm_rsize = fsp->fs_rtmax; } if (nmp->nm_rsize < NFS_FABLKSIZE) nmp->nm_rsize = NFS_FABLKSIZE; if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize) && fsp->fs_dtpref >= NFS_DIRBLKSIZ) nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) & ~(NFS_DIRBLKSIZ - 1); if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) { nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1); if (nmp->nm_readdirsize == 0) nmp->nm_readdirsize = fsp->fs_rtmax; } if (nmp->nm_readdirsize < NFS_DIRBLKSIZ) nmp->nm_readdirsize = NFS_DIRBLKSIZ; if (fsp->fs_maxfilesize > 0 && fsp->fs_maxfilesize < nmp->nm_maxfilesize) nmp->nm_maxfilesize = fsp->fs_maxfilesize; nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp); nmp->nm_state |= NFSSTA_GOTFSINFO; } /* * Lookups source address which should be used to communicate with * @nmp and stores it inside @pdst. * * Returns 0 on success. */ u_int8_t * nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p) { #if defined(INET6) || defined(INET) int error, fibnum; fibnum = curthread->td_proc->p_fibnum; #endif #ifdef INET if (nmp->nm_nam->sa_family == AF_INET) { struct sockaddr_in *sin; struct nhop4_extended nh_ext; sin = (struct sockaddr_in *)nmp->nm_nam; CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0, &nh_ext); CURVNET_RESTORE(); if (error != 0) return (NULL); if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { /* Ignore loopback addresses */ return (NULL); } *isinet6p = 0; *((struct in_addr *)paddr) = nh_ext.nh_src; return (u_int8_t *)paddr; } #endif #ifdef INET6 if (nmp->nm_nam->sa_family == AF_INET6) { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)nmp->nm_nam; CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred)); error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr, sin6->sin6_scope_id, NULL, paddr, NULL); CURVNET_RESTORE(); if (error != 0) return (NULL); if (IN6_IS_ADDR_LOOPBACK(paddr)) return (NULL); /* Scope is embedded in */ *isinet6p = 1; return (u_int8_t *)paddr; } #endif return (NULL); } /* * Copy NFS uid, gids from the cred structure. */ void newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr) { int i; KASSERT(cr->cr_ngroups >= 0, ("newnfs_copyincred: negative cr_ngroups")); nfscr->nfsc_uid = cr->cr_uid; nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1); for (i = 0; i < nfscr->nfsc_ngroups; i++) nfscr->nfsc_groups[i] = cr->cr_groups[i]; } /* * Do any client specific initialization. */ void nfscl_init(void) { static int inited = 0; if (inited) return; inited = 1; nfscl_inited = 1; ncl_pbuf_freecnt = nswbuf / 2 + 1; } /* * Check each of the attributes to be set, to ensure they aren't already * the correct value. Disable setting ones already correct. */ int nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap) { if (vap->va_mode != (mode_t)VNOVAL) { if (vap->va_mode == nvap->na_mode) vap->va_mode = (mode_t)VNOVAL; } if (vap->va_uid != (uid_t)VNOVAL) { if (vap->va_uid == nvap->na_uid) vap->va_uid = (uid_t)VNOVAL; } if (vap->va_gid != (gid_t)VNOVAL) { if (vap->va_gid == nvap->na_gid) vap->va_gid = (gid_t)VNOVAL; } if (vap->va_size != VNOVAL) { if (vap->va_size == nvap->na_size) vap->va_size = VNOVAL; } /* * We are normally called with only a partially initialized * VAP. Since the NFSv3 spec says that server may use the * file attributes to store the verifier, the spec requires * us to do a SETATTR RPC. FreeBSD servers store the verifier * in atime, but we can't really assume that all servers will * so we ensure that our SETATTR sets both atime and mtime. * Set the VA_UTIMES_NULL flag for this case, so that * the server's time will be used. This is needed to * work around a bug in some Solaris servers, where * setting the time TOCLIENT causes the Setattr RPC * to return NFS_OK, but not set va_mode. */ if (vap->va_mtime.tv_sec == VNOVAL) { vfs_timestamp(&vap->va_mtime); vap->va_vaflags |= VA_UTIMES_NULL; } if (vap->va_atime.tv_sec == VNOVAL) vap->va_atime = vap->va_mtime; return (1); } /* * Map nfsv4 errors to errno.h errors. * The uid and gid arguments are only used for NFSERR_BADOWNER and that * error should only be returned for the Open, Create and Setattr Ops. * As such, most calls can just pass in 0 for those arguments. */ APPLESTATIC int nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid) { struct proc *p; if (error < 10000 || error >= NFSERR_STALEWRITEVERF) return (error); if (td != NULL) p = td->td_proc; else p = NULL; switch (error) { case NFSERR_BADOWNER: tprintf(p, LOG_INFO, "No name and/or group mapping for uid,gid:(%d,%d)\n", uid, gid); return (EPERM); case NFSERR_BADNAME: case NFSERR_BADCHAR: printf("nfsv4 char/name not handled by server\n"); return (ENOENT); case NFSERR_STALECLIENTID: case NFSERR_STALESTATEID: case NFSERR_EXPIRED: case NFSERR_BADSTATEID: case NFSERR_BADSESSION: printf("nfsv4 recover err returned %d\n", error); return (EIO); case NFSERR_BADHANDLE: case NFSERR_SERVERFAULT: case NFSERR_BADTYPE: case NFSERR_FHEXPIRED: case NFSERR_RESOURCE: case NFSERR_MOVED: case NFSERR_NOFILEHANDLE: case NFSERR_MINORVERMISMATCH: case NFSERR_OLDSTATEID: case NFSERR_BADSEQID: case NFSERR_LEASEMOVED: case NFSERR_RECLAIMBAD: case NFSERR_BADXDR: case NFSERR_OPILLEGAL: printf("nfsv4 client/server protocol prob err=%d\n", error); return (EIO); default: tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error); return (EIO); }; } /* * Check to see if the process for this owner exists. Return 1 if it doesn't * and 0 otherwise. */ int nfscl_procdoesntexist(u_int8_t *own) { union { u_int32_t lval; u_int8_t cval[4]; } tl; struct proc *p; pid_t pid; int i, ret = 0; /* For the single open_owner of all 0 bytes, just return 0. */ for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++) if (own[i] != 0) break; if (i == NFSV4CL_LOCKNAMELEN) return (0); tl.cval[0] = *own++; tl.cval[1] = *own++; tl.cval[2] = *own++; tl.cval[3] = *own++; pid = tl.lval; p = pfind_locked(pid); if (p == NULL) return (1); if (p->p_stats == NULL) { PROC_UNLOCK(p); return (0); } tl.cval[0] = *own++; tl.cval[1] = *own++; tl.cval[2] = *own++; tl.cval[3] = *own++; if (tl.lval != p->p_stats->p_start.tv_sec) { ret = 1; } else { tl.cval[0] = *own++; tl.cval[1] = *own++; tl.cval[2] = *own++; tl.cval[3] = *own; if (tl.lval != p->p_stats->p_start.tv_usec) ret = 1; } PROC_UNLOCK(p); return (ret); } /* * - nfs pseudo system call for the client */ /* * MPSAFE */ static int nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap) { struct file *fp; struct nfscbd_args nfscbdarg; struct nfsd_nfscbd_args nfscbdarg2; struct nameidata nd; struct nfscl_dumpmntopts dumpmntopts; cap_rights_t rights; char *buf; int error; if (uap->flag & NFSSVC_CBADDSOCK) { error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg)); if (error) return (error); /* * 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, nfscbdarg.sock, cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp); if (error) return (error); if (fp->f_type != DTYPE_SOCKET) { fdrop(fp, td); return (EPERM); } error = nfscbd_addsock(fp); fdrop(fp, td); if (!error && nfscl_enablecallb == 0) { nfsv4_cbport = nfscbdarg.port; nfscl_enablecallb = 1; } } else if (uap->flag & NFSSVC_NFSCBD) { if (uap->argp == NULL) return (EINVAL); error = copyin(uap->argp, (caddr_t)&nfscbdarg2, sizeof(nfscbdarg2)); if (error) return (error); error = nfscbd_nfsd(td, &nfscbdarg2); } else if (uap->flag & NFSSVC_DUMPMNTOPTS) { error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts)); if (error == 0 && (dumpmntopts.ndmnt_blen < 256 || dumpmntopts.ndmnt_blen > 1024)) error = EINVAL; if (error == 0) error = nfsrv_lookupfilename(&nd, dumpmntopts.ndmnt_fname, td); if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { vput(nd.ni_vp); error = EINVAL; } if (error == 0) { buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK); nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf, dumpmntopts.ndmnt_blen); vput(nd.ni_vp); error = copyout(buf, dumpmntopts.ndmnt_buf, dumpmntopts.ndmnt_blen); free(buf, M_TEMP); } } else { error = EINVAL; } return (error); } extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *); /* * Called once to initialize data structures... */ static int nfscl_modevent(module_t mod, int type, void *data) { int error = 0; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded) return (0); newnfs_portinit(); mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF); nfscl_init(); NFSD_LOCK(); nfsrvd_cbinit(0); NFSD_UNLOCK(); ncl_call_invalcaches = ncl_invalcaches; nfsd_call_nfscl = nfssvc_nfscl; loaded = 1; break; case MOD_UNLOAD: if (nfs_numnfscbd != 0) { error = EBUSY; break; } /* * XXX: Unloading of nfscl module is unsupported. */ #if 0 ncl_call_invalcaches = NULL; nfsd_call_nfscl = NULL; /* and get rid of the mutexes */ mtx_destroy(&ncl_iod_mutex); loaded = 0; break; #else /* FALLTHROUGH */ #endif default: error = EOPNOTSUPP; break; } return error; } static moduledata_t nfscl_mod = { "nfscl", nfscl_modevent, NULL, }; DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfscl, 1); MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1); MODULE_DEPEND(nfscl, krpc, 1, 1, 1); MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1); MODULE_DEPEND(nfscl, nfslock, 1, 1, 1); Index: head/sys/fs/nfsclient/nfs_clstate.c =================================================================== --- head/sys/fs/nfsclient/nfs_clstate.c (revision 321627) +++ head/sys/fs/nfsclient/nfs_clstate.c (revision 321628) @@ -1,5318 +1,5317 @@ /*- * Copyright (c) 2009 Rick Macklem, University of Guelph * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); /* * These functions implement the client side state handling for NFSv4. * NFSv4 state handling: * - A lockowner is used to determine lock contention, so it * corresponds directly to a Posix pid. (1 to 1 mapping) * - The correct granularity of an OpenOwner is not nearly so * obvious. An OpenOwner does the following: * - provides a serial sequencing of Open/Close/Lock-with-new-lockowner * - is used to check for Open/Share contention (not applicable to * this client, since all Opens are Deny_None) * As such, I considered both extreme. * 1 OpenOwner per ClientID - Simple to manage, but fully serializes * all Open, Close and Lock (with a new lockowner) Ops. * 1 OpenOwner for each Open - This one results in an OpenConfirm for * every Open, for most servers. * So, I chose to use the same mapping as I did for LockOwnwers. * The main concern here is that you can end up with multiple Opens * for the same File Handle, but on different OpenOwners (opens * inherited from parents, grandparents...) and you do not know * which of these the vnodeop close applies to. This is handled by * delaying the Close Op(s) until all of the Opens have been closed. * (It is not yet obvious if this is the correct granularity.) * - How the code handles serialization: * - For the ClientId, it uses an exclusive lock while getting its * SetClientId and during recovery. Otherwise, it uses a shared * lock via a reference count. * - For the rest of the data structures, it uses an SMP mutex * (once the nfs client is SMP safe) and doesn't sleep while * manipulating the linked lists. * - The serialization of Open/Close/Lock/LockU falls out in the * "wash", since OpenOwners and LockOwners are both mapped from * Posix pid. In other words, there is only one Posix pid using * any given owner, so that owner is serialized. (If you change * the granularity of the OpenOwner, then code must be added to * serialize Ops on the OpenOwner.) * - When to get rid of OpenOwners and LockOwners. * - The function nfscl_cleanup_common() is executed after a process exits. * It goes through the client list looking for all Open and Lock Owners. * When one is found, it is marked "defunct" or in the case of * an OpenOwner without any Opens, freed. * The renew thread scans for defunct Owners and gets rid of them, * if it can. The LockOwners will also be deleted when the * associated Open is closed. * - If the LockU or Close Op(s) fail during close in a way * that could be recovered upon retry, they are relinked to the * ClientId's defunct open list and retried by the renew thread * until they succeed or an unmount/recovery occurs. * (Since we are done with them, they do not need to be recovered.) */ #ifndef APPLEKEXT #include /* * Global variables */ extern struct nfsstatsv1 nfsstatsv1; extern struct nfsreqhead nfsd_reqq; extern u_int32_t newnfs_false, newnfs_true; extern int nfscl_debuglevel; extern int nfscl_enablecallb; extern int nfs_numnfscbd; NFSREQSPINLOCK; NFSCLSTATEMUTEX; int nfscl_inited = 0; struct nfsclhead nfsclhead; /* Head of clientid list */ int nfscl_deleghighwater = NFSCLDELEGHIGHWATER; int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER; #endif /* !APPLEKEXT */ static int nfscl_delegcnt = 0; static int nfscl_layoutcnt = 0; static int nfscl_getopen(struct nfsclownerhead *, u_int8_t *, int, u_int8_t *, u_int8_t *, u_int32_t, struct nfscllockowner **, struct nfsclopen **); static void nfscl_clrelease(struct nfsclclient *); static void nfscl_cleanclient(struct nfsclclient *); static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *, struct ucred *, NFSPROC_T *); static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *, struct nfsmount *, struct ucred *, NFSPROC_T *); static void nfscl_recover(struct nfsclclient *, struct ucred *, NFSPROC_T *); static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *, struct nfscllock *, int); static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **, struct nfscllock **, int); static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *); static u_int32_t nfscl_nextcbident(void); static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **); static struct nfsclclient *nfscl_getclnt(u_int32_t); static struct nfsclclient *nfscl_getclntsess(uint8_t *); static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *, int); static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *, int, struct nfsclrecalllayout **); static void nfscl_reldevinfo_locked(struct nfscldevinfo *); static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *, int); static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *); static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *, u_int8_t *, struct nfscllock **); static void nfscl_freealllocks(struct nfscllockownerhead *, int); static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int, struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **); static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *, struct nfsclowner **, struct nfsclowner **, struct nfsclopen **, struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *); static int nfscl_moveopen(vnode_t , struct nfsclclient *, struct nfsmount *, struct nfsclopen *, struct nfsclowner *, struct nfscldeleg *, struct ucred *, NFSPROC_T *); static void nfscl_totalrecall(struct nfsclclient *); static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *, struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *); static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int, u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int, struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *); static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *, int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short, struct ucred *, NFSPROC_T *); static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t, struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *); static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *); static int nfscl_errmap(struct nfsrv_descript *, u_int32_t); static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *); static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *, struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int); static void nfscl_freeopenowner(struct nfsclowner *, int); static void nfscl_cleandeleg(struct nfscldeleg *); static int nfscl_trydelegreturn(struct nfscldeleg *, struct ucred *, struct nfsmount *, NFSPROC_T *); static void nfscl_emptylockowner(struct nfscllockowner *, struct nfscllockownerfhhead *); static void nfscl_mergeflayouts(struct nfsclflayouthead *, struct nfsclflayouthead *); static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t, uint64_t, uint32_t, struct nfsclrecalllayout *); static int nfscl_seq(uint32_t, uint32_t); static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *, struct ucred *, NFSPROC_T *); static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *, struct ucred *, NFSPROC_T *); static short nfscberr_null[] = { 0, 0, }; static short nfscberr_getattr[] = { NFSERR_RESOURCE, NFSERR_BADHANDLE, NFSERR_BADXDR, NFSERR_RESOURCE, NFSERR_SERVERFAULT, 0, }; static short nfscberr_recall[] = { NFSERR_RESOURCE, NFSERR_BADHANDLE, NFSERR_BADSTATEID, NFSERR_BADXDR, NFSERR_RESOURCE, NFSERR_SERVERFAULT, 0, }; static short *nfscl_cberrmap[] = { nfscberr_null, nfscberr_null, nfscberr_null, nfscberr_getattr, nfscberr_recall }; #define NETFAMILY(clp) \ (((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET) /* * Called for an open operation. * If the nfhp argument is NULL, just get an openowner. */ APPLESTATIC int nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg, struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp, struct nfsclopen **opp, int *newonep, int *retp, int lockit) { struct nfsclclient *clp; struct nfsclowner *owp, *nowp; struct nfsclopen *op = NULL, *nop = NULL; struct nfscldeleg *dp; struct nfsclownerhead *ohp; u_int8_t own[NFSV4CL_LOCKNAMELEN]; int ret; if (newonep != NULL) *newonep = 0; if (opp != NULL) *opp = NULL; if (owpp != NULL) *owpp = NULL; /* * Might need one or both of these, so MALLOC them now, to * avoid a tsleep() in MALLOC later. */ MALLOC(nowp, struct nfsclowner *, sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK); if (nfhp != NULL) MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) + fhlen - 1, M_NFSCLOPEN, M_WAITOK); ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp); if (ret != 0) { FREE((caddr_t)nowp, M_NFSCLOWNER); if (nop != NULL) FREE((caddr_t)nop, M_NFSCLOPEN); return (ret); } /* * Get the Open iff it already exists. * If none found, add the new one or return error, depending upon * "create". */ NFSLOCKCLSTATE(); dp = NULL; /* First check the delegation list */ if (nfhp != NULL && usedeleg) { LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) { if (dp->nfsdl_fhlen == fhlen && !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) { if (!(amode & NFSV4OPEN_ACCESSWRITE) || (dp->nfsdl_flags & NFSCLDL_WRITE)) break; dp = NULL; break; } } } if (dp != NULL) { nfscl_filllockowner(p->td_proc, own, F_POSIX); ohp = &dp->nfsdl_owner; } else { /* For NFSv4.1 and this option, use a single open_owner. */ if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) nfscl_filllockowner(NULL, own, F_POSIX); else nfscl_filllockowner(p->td_proc, own, F_POSIX); ohp = &clp->nfsc_owner; } /* Now, search for an openowner */ LIST_FOREACH(owp, ohp, nfsow_list) { if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN)) break; } /* * Create a new open, as required. */ nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen, cred, newonep); /* * Now, check the mode on the open and return the appropriate * value. */ if (retp != NULL) { if (nfhp != NULL && dp != NULL && nop == NULL) /* new local open on delegation */ *retp = NFSCLOPEN_SETCRED; else *retp = NFSCLOPEN_OK; } if (op != NULL && (amode & ~(op->nfso_mode))) { op->nfso_mode |= amode; if (retp != NULL && dp == NULL) *retp = NFSCLOPEN_DOOPEN; } /* * Serialize modifications to the open owner for multiple threads * within the same process using a read/write sleep lock. * For NFSv4.1 and a single OpenOwner, allow concurrent open operations * by acquiring a shared lock. The close operations still use an * exclusive lock for this case. */ if (lockit != 0) { if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) { /* * Get a shared lock on the OpenOwner, but first * wait for any pending exclusive lock, so that the * exclusive locker gets priority. */ nfsv4_lock(&owp->nfsow_rwlock, 0, NULL, NFSCLSTATEMUTEXPTR, NULL); nfsv4_getref(&owp->nfsow_rwlock, NULL, NFSCLSTATEMUTEXPTR, NULL); } else nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR); } NFSUNLOCKCLSTATE(); if (nowp != NULL) FREE((caddr_t)nowp, M_NFSCLOWNER); if (nop != NULL) FREE((caddr_t)nop, M_NFSCLOPEN); if (owpp != NULL) *owpp = owp; if (opp != NULL) *opp = op; return (0); } /* * Create a new open, as required. */ static void nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp, struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp, struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen, struct ucred *cred, int *newonep) { struct nfsclowner *owp = *owpp, *nowp; struct nfsclopen *op, *nop; if (nowpp != NULL) nowp = *nowpp; else nowp = NULL; if (nopp != NULL) nop = *nopp; else nop = NULL; if (owp == NULL && nowp != NULL) { NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN); LIST_INIT(&nowp->nfsow_open); nowp->nfsow_clp = clp; nowp->nfsow_seqid = 0; nowp->nfsow_defunct = 0; nfscl_lockinit(&nowp->nfsow_rwlock); if (dp != NULL) { nfsstatsv1.cllocalopenowners++; LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list); } else { nfsstatsv1.clopenowners++; LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list); } owp = *owpp = nowp; *nowpp = NULL; if (newonep != NULL) *newonep = 1; } /* If an fhp has been specified, create an Open as well. */ if (fhp != NULL) { /* and look for the correct open, based upon FH */ LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == fhlen && !NFSBCMP(op->nfso_fh, fhp, fhlen)) break; } if (op == NULL && nop != NULL) { nop->nfso_own = owp; nop->nfso_mode = 0; nop->nfso_opencnt = 0; nop->nfso_posixlock = 1; nop->nfso_fhlen = fhlen; NFSBCOPY(fhp, nop->nfso_fh, fhlen); LIST_INIT(&nop->nfso_lock); nop->nfso_stateid.seqid = 0; nop->nfso_stateid.other[0] = 0; nop->nfso_stateid.other[1] = 0; nop->nfso_stateid.other[2] = 0; KASSERT(cred != NULL, ("%s: cred NULL\n", __func__)); newnfs_copyincred(cred, &nop->nfso_cred); if (dp != NULL) { TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list); TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list); dp->nfsdl_timestamp = NFSD_MONOSEC + 120; nfsstatsv1.cllocalopens++; } else { nfsstatsv1.clopens++; } LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list); *opp = nop; *nopp = NULL; if (newonep != NULL) *newonep = 1; } else { *opp = op; } } } /* * Called to find/add a delegation to a client. */ APPLESTATIC int nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp, int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp) { struct nfscldeleg *dp = *dpp, *tdp; /* * First, if we have received a Read delegation for a file on a * read/write file system, just return it, because they aren't * useful, imho. */ if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) && (dp->nfsdl_flags & NFSCLDL_READ)) { (void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p); FREE((caddr_t)dp, M_NFSCLDELEG); *dpp = NULL; return (0); } /* Look for the correct deleg, based upon FH */ NFSLOCKCLSTATE(); tdp = nfscl_finddeleg(clp, nfhp, fhlen); if (tdp == NULL) { if (dp == NULL) { NFSUNLOCKCLSTATE(); return (NFSERR_BADSTATEID); } *dpp = NULL; TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list); LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp, nfsdl_hash); dp->nfsdl_timestamp = NFSD_MONOSEC + 120; nfsstatsv1.cldelegates++; nfscl_delegcnt++; } else { /* * Delegation already exists, what do we do if a new one?? */ if (dp != NULL) { printf("Deleg already exists!\n"); FREE((caddr_t)dp, M_NFSCLDELEG); *dpp = NULL; } else { *dpp = tdp; } } NFSUNLOCKCLSTATE(); return (0); } /* * Find a delegation for this file handle. Return NULL upon failure. */ static struct nfscldeleg * nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen) { struct nfscldeleg *dp; LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) { if (dp->nfsdl_fhlen == fhlen && !NFSBCMP(dp->nfsdl_fh, fhp, fhlen)) break; } return (dp); } /* * Get a stateid for an I/O operation. First, look for an open and iff * found, return either a lockowner stateid or the open stateid. * If no Open is found, just return error and the special stateid of all zeros. */ APPLESTATIC int nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode, int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp, void **lckpp) { struct nfsclclient *clp; struct nfsclowner *owp; struct nfsclopen *op = NULL, *top; struct nfscllockowner *lp; struct nfscldeleg *dp; struct nfsnode *np; struct nfsmount *nmp; u_int8_t own[NFSV4CL_LOCKNAMELEN]; int error, done; *lckpp = NULL; /* * Initially, just set the special stateid of all zeros. * (Don't do this for a DS, since the special stateid can't be used.) */ if (fords == 0) { stateidp->seqid = 0; stateidp->other[0] = 0; stateidp->other[1] = 0; stateidp->other[2] = 0; } if (vnode_vtype(vp) != VREG) return (EISDIR); np = VTONFS(vp); nmp = VFSTONFS(vnode_mount(vp)); NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return (EACCES); } /* * Wait for recovery to complete. */ while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG)) (void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR, PZERO, "nfsrecvr", NULL); /* * First, look for a delegation. */ LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) { if (dp->nfsdl_fhlen == fhlen && !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) { if (!(mode & NFSV4OPEN_ACCESSWRITE) || (dp->nfsdl_flags & NFSCLDL_WRITE)) { stateidp->seqid = dp->nfsdl_stateid.seqid; stateidp->other[0] = dp->nfsdl_stateid.other[0]; stateidp->other[1] = dp->nfsdl_stateid.other[1]; stateidp->other[2] = dp->nfsdl_stateid.other[2]; if (!(np->n_flag & NDELEGRECALL)) { TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list); TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list); dp->nfsdl_timestamp = NFSD_MONOSEC + 120; dp->nfsdl_rwlock.nfslock_usecnt++; *lckpp = (void *)&dp->nfsdl_rwlock; } NFSUNLOCKCLSTATE(); return (0); } break; } } if (p != NULL) { /* * If p != NULL, we want to search the parentage tree * for a matching OpenOwner and use that. */ if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) nfscl_filllockowner(NULL, own, F_POSIX); else nfscl_filllockowner(p->td_proc, own, F_POSIX); lp = NULL; error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own, mode, &lp, &op); if (error == 0 && lp != NULL && fords == 0) { /* Don't return a lock stateid for a DS. */ stateidp->seqid = lp->nfsl_stateid.seqid; stateidp->other[0] = lp->nfsl_stateid.other[0]; stateidp->other[1] = lp->nfsl_stateid.other[1]; stateidp->other[2] = lp->nfsl_stateid.other[2]; NFSUNLOCKCLSTATE(); return (0); } } if (op == NULL) { /* If not found, just look for any OpenOwner that will work. */ top = NULL; done = 0; owp = LIST_FIRST(&clp->nfsc_owner); while (!done && owp != NULL) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == fhlen && !NFSBCMP(op->nfso_fh, nfhp, fhlen)) { if (top == NULL && (op->nfso_mode & NFSV4OPEN_ACCESSWRITE) != 0 && (mode & NFSV4OPEN_ACCESSREAD) != 0) top = op; if ((mode & op->nfso_mode) == mode) { done = 1; break; } } } if (!done) owp = LIST_NEXT(owp, nfsow_list); } if (!done) { NFSCL_DEBUG(2, "openmode top=%p\n", top); if (top == NULL || NFSHASOPENMODE(nmp)) { NFSUNLOCKCLSTATE(); return (ENOENT); } else op = top; } /* * For read aheads or write behinds, use the open cred. * A read ahead or write behind is indicated by p == NULL. */ if (p == NULL) newnfs_copycred(&op->nfso_cred, cred); } /* * No lock stateid, so return the open stateid. */ stateidp->seqid = op->nfso_stateid.seqid; stateidp->other[0] = op->nfso_stateid.other[0]; stateidp->other[1] = op->nfso_stateid.other[1]; stateidp->other[2] = op->nfso_stateid.other[2]; NFSUNLOCKCLSTATE(); return (0); } /* * Search for a matching file, mode and, optionally, lockowner. */ static int nfscl_getopen(struct nfsclownerhead *ohp, u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown, u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp) { struct nfsclowner *owp; struct nfsclopen *op, *rop, *rop2; struct nfscllockowner *lp; int keep_looping; if (lpp != NULL) *lpp = NULL; /* * rop will be set to the open to be returned. There are three * variants of this, all for an open of the correct file: * 1 - A match of lockown. * 2 - A match of the openown, when no lockown match exists. * 3 - A match for any open, if no openown or lockown match exists. * Looking for #2 over #3 probably isn't necessary, but since * RFC3530 is vague w.r.t. the relationship between openowners and * lockowners, I think this is the safer way to go. */ rop = NULL; rop2 = NULL; keep_looping = 1; /* Search the client list */ owp = LIST_FIRST(ohp); while (owp != NULL && keep_looping != 0) { /* and look for the correct open */ op = LIST_FIRST(&owp->nfsow_open); while (op != NULL && keep_looping != 0) { if (op->nfso_fhlen == fhlen && !NFSBCMP(op->nfso_fh, nfhp, fhlen) && (op->nfso_mode & mode) == mode) { if (lpp != NULL) { /* Now look for a matching lockowner. */ LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (!NFSBCMP(lp->nfsl_owner, lockown, NFSV4CL_LOCKNAMELEN)) { *lpp = lp; rop = op; keep_looping = 0; break; } } } if (rop == NULL && !NFSBCMP(owp->nfsow_owner, openown, NFSV4CL_LOCKNAMELEN)) { rop = op; if (lpp == NULL) keep_looping = 0; } if (rop2 == NULL) rop2 = op; } op = LIST_NEXT(op, nfso_list); } owp = LIST_NEXT(owp, nfsow_list); } if (rop == NULL) rop = rop2; if (rop == NULL) return (EBADF); *opp = rop; return (0); } /* * Release use of an open owner. Called when open operations are done * with the open owner. */ APPLESTATIC void nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp, __unused int error, __unused int candelete, int unlocked) { if (owp == NULL) return; NFSLOCKCLSTATE(); if (unlocked == 0) { if (NFSHASONEOPENOWN(nmp)) nfsv4_relref(&owp->nfsow_rwlock); else nfscl_lockunlock(&owp->nfsow_rwlock); } nfscl_clrelease(owp->nfsow_clp); NFSUNLOCKCLSTATE(); } /* * Release use of an open structure under an open owner. */ APPLESTATIC void nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error, int candelete) { struct nfsclclient *clp; struct nfsclowner *owp; if (op == NULL) return; NFSLOCKCLSTATE(); owp = op->nfso_own; if (NFSHASONEOPENOWN(nmp)) nfsv4_relref(&owp->nfsow_rwlock); else nfscl_lockunlock(&owp->nfsow_rwlock); clp = owp->nfsow_clp; if (error && candelete && op->nfso_opencnt == 0) nfscl_freeopen(op, 0); nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); } /* * Called to get a clientid structure. It will optionally lock the * client data structures to do the SetClientId/SetClientId_confirm, * but will release that lock and return the clientid with a reference * count on it. * If the "cred" argument is NULL, a new clientid should not be created. * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot * be done. * The start_renewthread argument tells nfscl_getcl() to start a renew * thread if this creates a new clp. * It always clpp with a reference count on it, unless returning an error. */ APPLESTATIC int nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p, int start_renewthread, struct nfsclclient **clpp) { struct nfsclclient *clp; struct nfsclclient *newclp = NULL; struct nfsmount *nmp; char uuid[HOSTUUIDLEN]; int igotlock = 0, error, trystalecnt, clidinusedelay, i; u_int16_t idlen = 0; nmp = VFSTONFS(mp); if (cred != NULL) { getcredhostuuid(cred, uuid, sizeof uuid); idlen = strlen(uuid); if (idlen > 0) idlen += sizeof (u_int64_t); else idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */ MALLOC(newclp, struct nfsclclient *, sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT, M_WAITOK | M_ZERO); } NFSLOCKCLSTATE(); /* * If a forced dismount is already in progress, don't * allocate a new clientid and get out now. For the case where * clp != NULL, this is a harmless optimization. */ - if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (NFSCL_FORCEDISM(mp)) { NFSUNLOCKCLSTATE(); if (newclp != NULL) free(newclp, M_NFSCLCLIENT); return (EBADF); } clp = nmp->nm_clp; if (clp == NULL) { if (newclp == NULL) { NFSUNLOCKCLSTATE(); return (EACCES); } clp = newclp; clp->nfsc_idlen = idlen; LIST_INIT(&clp->nfsc_owner); TAILQ_INIT(&clp->nfsc_deleg); TAILQ_INIT(&clp->nfsc_layout); LIST_INIT(&clp->nfsc_devinfo); for (i = 0; i < NFSCLDELEGHASHSIZE; i++) LIST_INIT(&clp->nfsc_deleghash[i]); for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++) LIST_INIT(&clp->nfsc_layouthash[i]); clp->nfsc_flags = NFSCLFLAGS_INITED; clp->nfsc_clientidrev = 1; clp->nfsc_cbident = nfscl_nextcbident(); nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id, clp->nfsc_idlen); LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list); nmp->nm_clp = clp; clp->nfsc_nmp = nmp; NFSUNLOCKCLSTATE(); if (start_renewthread != 0) nfscl_start_renewthread(clp); } else { NFSUNLOCKCLSTATE(); if (newclp != NULL) free(newclp, M_NFSCLCLIENT); } NFSLOCKCLSTATE(); while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock && - (mp->mnt_kern_flag & MNTK_UNMOUNTF) == 0) + !NFSCL_FORCEDISM(mp)) igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL, NFSCLSTATEMUTEXPTR, mp); if (igotlock == 0) { /* * Call nfsv4_lock() with "iwantlock == 0" so that it will * wait for a pending exclusive lock request. This gives the * exclusive lock request priority over this shared lock * request. * An exclusive lock on nfsc_lock is used mainly for server * crash recoveries. */ nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp); nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp); } - if (igotlock == 0 && (mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (igotlock == 0 && NFSCL_FORCEDISM(mp)) { /* * Both nfsv4_lock() and nfsv4_getref() know to check - * for MNTK_UNMOUNTF and return without sleeping to + * for NFSCL_FORCEDISM() and return without sleeping to * wait for the exclusive lock to be released, since it * might be held by nfscl_umount() and we need to get out * now for that case and not wait until nfscl_umount() * releases it. */ NFSUNLOCKCLSTATE(); return (EBADF); } NFSUNLOCKCLSTATE(); /* * If it needs a clientid, do the setclientid now. */ if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) { if (!igotlock) panic("nfscl_clget"); if (p == NULL || cred == NULL) { NFSLOCKCLSTATE(); nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); return (EACCES); } /* * If RFC3530 Sec. 14.2.33 is taken literally, * NFSERR_CLIDINUSE will be returned persistently for the * case where a new mount of the same file system is using * a different principal. In practice, NFSERR_CLIDINUSE is * only returned when there is outstanding unexpired state * on the clientid. As such, try for twice the lease * interval, if we know what that is. Otherwise, make a * wild ass guess. * The case of returning NFSERR_STALECLIENTID is far less * likely, but might occur if there is a significant delay * between doing the SetClientID and SetClientIDConfirm Ops, * such that the server throws away the clientid before * receiving the SetClientIDConfirm. */ if (clp->nfsc_renew > 0) clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2; else clidinusedelay = 120; trystalecnt = 3; do { error = nfsrpc_setclient(nmp, clp, 0, cred, p); if (error == NFSERR_STALECLIENTID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_BADSESSION || error == NFSERR_CLIDINUSE) { (void) nfs_catnap(PZERO, error, "nfs_setcl"); } } while (((error == NFSERR_STALECLIENTID || error == NFSERR_BADSESSION || error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) || (error == NFSERR_CLIDINUSE && --clidinusedelay > 0)); if (error) { NFSLOCKCLSTATE(); nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); return (error); } clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID; } if (igotlock) { NFSLOCKCLSTATE(); nfsv4_unlock(&clp->nfsc_lock, 1); NFSUNLOCKCLSTATE(); } *clpp = clp; return (0); } /* * Get a reference to a clientid and return it, if valid. */ APPLESTATIC struct nfsclclient * nfscl_findcl(struct nfsmount *nmp) { struct nfsclclient *clp; clp = nmp->nm_clp; if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) return (NULL); return (clp); } /* * Release the clientid structure. It may be locked or reference counted. */ static void nfscl_clrelease(struct nfsclclient *clp) { if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK) nfsv4_unlock(&clp->nfsc_lock, 0); else nfsv4_relref(&clp->nfsc_lock); } /* * External call for nfscl_clrelease. */ APPLESTATIC void nfscl_clientrelease(struct nfsclclient *clp) { NFSLOCKCLSTATE(); if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK) nfsv4_unlock(&clp->nfsc_lock, 0); else nfsv4_relref(&clp->nfsc_lock); NFSUNLOCKCLSTATE(); } /* * Called when wanting to lock a byte region. */ APPLESTATIC int nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp, int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp, struct nfscllockowner **lpp, int *newonep, int *donelocallyp) { struct nfscllockowner *lp; struct nfsclopen *op; struct nfsclclient *clp; struct nfscllockowner *nlp; struct nfscllock *nlop, *otherlop; struct nfscldeleg *dp = NULL, *ldp = NULL; struct nfscllockownerhead *lhp = NULL; struct nfsnode *np; u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN]; u_int8_t *openownp; int error = 0, ret, donelocally = 0; u_int32_t mode; /* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */ mode = 0; np = VTONFS(vp); *lpp = NULL; lp = NULL; *newonep = 0; *donelocallyp = 0; /* * Might need these, so MALLOC them now, to * avoid a tsleep() in MALLOC later. */ MALLOC(nlp, struct nfscllockowner *, sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK); MALLOC(otherlop, struct nfscllock *, sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK); MALLOC(nlop, struct nfscllock *, sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK); nlop->nfslo_type = type; nlop->nfslo_first = off; if (len == NFS64BITSSET) { nlop->nfslo_end = NFS64BITSSET; } else { nlop->nfslo_end = off + len; if (nlop->nfslo_end <= nlop->nfslo_first) error = NFSERR_INVAL; } if (!error) { if (recovery) clp = rclp; else error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp); } if (error) { FREE((caddr_t)nlp, M_NFSCLLOCKOWNER); FREE((caddr_t)otherlop, M_NFSCLLOCK); FREE((caddr_t)nlop, M_NFSCLLOCK); return (error); } op = NULL; if (recovery) { ownp = rownp; openownp = ropenownp; } else { nfscl_filllockowner(id, own, flags); ownp = own; if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp)))) nfscl_filllockowner(NULL, openown, F_POSIX); else nfscl_filllockowner(p->td_proc, openown, F_POSIX); openownp = openown; } if (!recovery) { NFSLOCKCLSTATE(); /* * First, search for a delegation. If one exists for this file, * the lock can be done locally against it, so long as there * isn't a local lock conflict. */ ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); /* Just sanity check for correct type of delegation */ if (dp != NULL && ((dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 || (type == F_WRLCK && (dp->nfsdl_flags & NFSCLDL_WRITE) == 0))) dp = NULL; } if (dp != NULL) { /* Now, find an open and maybe a lockowner. */ ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op); if (ret) ret = nfscl_getopen(&clp->nfsc_owner, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op); if (!ret) { lhp = &dp->nfsdl_lock; TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list); TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list); dp->nfsdl_timestamp = NFSD_MONOSEC + 120; donelocally = 1; } else { dp = NULL; } } if (!donelocally) { /* * Get the related Open and maybe lockowner. */ error = nfscl_getopen(&clp->nfsc_owner, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp, ownp, mode, &lp, &op); if (!error) lhp = &op->nfso_lock; } if (!error && !recovery) error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, nlop, ownp, ldp, NULL); if (error) { if (!recovery) { nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); } FREE((caddr_t)nlp, M_NFSCLLOCKOWNER); FREE((caddr_t)otherlop, M_NFSCLLOCK); FREE((caddr_t)nlop, M_NFSCLLOCK); return (error); } /* * Ok, see if a lockowner exists and create one, as required. */ if (lp == NULL) LIST_FOREACH(lp, lhp, nfsl_list) { if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN)) break; } if (lp == NULL) { NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN); if (recovery) NFSBCOPY(ropenownp, nlp->nfsl_openowner, NFSV4CL_LOCKNAMELEN); else NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner, NFSV4CL_LOCKNAMELEN); nlp->nfsl_seqid = 0; nlp->nfsl_lockflags = flags; nlp->nfsl_inprog = NULL; nfscl_lockinit(&nlp->nfsl_rwlock); LIST_INIT(&nlp->nfsl_lock); if (donelocally) { nlp->nfsl_open = NULL; nfsstatsv1.cllocallockowners++; } else { nlp->nfsl_open = op; nfsstatsv1.cllockowners++; } LIST_INSERT_HEAD(lhp, nlp, nfsl_list); lp = nlp; nlp = NULL; *newonep = 1; } /* * Now, update the byte ranges for locks. */ ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally); if (!ret) donelocally = 1; if (donelocally) { *donelocallyp = 1; if (!recovery) nfscl_clrelease(clp); } else { /* * Serial modifications on the lock owner for multiple threads * for the same process using a read/write lock. */ if (!recovery) nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR); } if (!recovery) NFSUNLOCKCLSTATE(); if (nlp) FREE((caddr_t)nlp, M_NFSCLLOCKOWNER); if (nlop) FREE((caddr_t)nlop, M_NFSCLLOCK); if (otherlop) FREE((caddr_t)otherlop, M_NFSCLLOCK); *lpp = lp; return (0); } /* * Called to unlock a byte range, for LockU. */ APPLESTATIC int nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len, __unused struct ucred *cred, NFSPROC_T *p, int callcnt, struct nfsclclient *clp, void *id, int flags, struct nfscllockowner **lpp, int *dorpcp) { struct nfscllockowner *lp; struct nfsclowner *owp; struct nfsclopen *op; struct nfscllock *nlop, *other_lop = NULL; struct nfscldeleg *dp; struct nfsnode *np; u_int8_t own[NFSV4CL_LOCKNAMELEN]; int ret = 0, fnd; np = VTONFS(vp); *lpp = NULL; *dorpcp = 0; /* * Might need these, so MALLOC them now, to * avoid a tsleep() in MALLOC later. */ MALLOC(nlop, struct nfscllock *, sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK); nlop->nfslo_type = F_UNLCK; nlop->nfslo_first = off; if (len == NFS64BITSSET) { nlop->nfslo_end = NFS64BITSSET; } else { nlop->nfslo_end = off + len; if (nlop->nfslo_end <= nlop->nfslo_first) { FREE((caddr_t)nlop, M_NFSCLLOCK); return (NFSERR_INVAL); } } if (callcnt == 0) { MALLOC(other_lop, struct nfscllock *, sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK); *other_lop = *nlop; } nfscl_filllockowner(id, own, flags); dp = NULL; NFSLOCKCLSTATE(); if (callcnt == 0) dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); /* * First, unlock any local regions on a delegation. */ if (dp != NULL) { /* Look for this lockowner. */ LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) break; } if (lp != NULL) /* Use other_lop, so nlop is still available */ (void)nfscl_updatelock(lp, &other_lop, NULL, 1); } /* * Now, find a matching open/lockowner that hasn't already been done, * as marked by nfsl_inprog. */ lp = NULL; fnd = 0; LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == np->n_fhp->nfh_len && !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) { LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (lp->nfsl_inprog == NULL && !NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) { fnd = 1; break; } } if (fnd) break; } } if (fnd) break; } if (lp != NULL) { ret = nfscl_updatelock(lp, &nlop, NULL, 0); if (ret) *dorpcp = 1; /* * Serial modifications on the lock owner for multiple * threads for the same process using a read/write lock. */ lp->nfsl_inprog = p; nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR); *lpp = lp; } NFSUNLOCKCLSTATE(); if (nlop) FREE((caddr_t)nlop, M_NFSCLLOCK); if (other_lop) FREE((caddr_t)other_lop, M_NFSCLLOCK); return (0); } /* * Release all lockowners marked in progess for this process and file. */ APPLESTATIC void nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p, void *id, int flags) { struct nfsclowner *owp; struct nfsclopen *op; struct nfscllockowner *lp; struct nfsnode *np; u_int8_t own[NFSV4CL_LOCKNAMELEN]; np = VTONFS(vp); nfscl_filllockowner(id, own, flags); NFSLOCKCLSTATE(); LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == np->n_fhp->nfh_len && !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) { LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (lp->nfsl_inprog == p && !NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) { lp->nfsl_inprog = NULL; nfscl_lockunlock(&lp->nfsl_rwlock); } } } } } nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); } /* * Called to find out if any bytes within the byte range specified are * write locked by the calling process. Used to determine if flushing * is required before a LockU. * If in doubt, return 1, so the flush will occur. */ APPLESTATIC int nfscl_checkwritelocked(vnode_t vp, struct flock *fl, struct ucred *cred, NFSPROC_T *p, void *id, int flags) { struct nfsclowner *owp; struct nfscllockowner *lp; struct nfsclopen *op; struct nfsclclient *clp; struct nfscllock *lop; struct nfscldeleg *dp; struct nfsnode *np; u_int64_t off, end; u_int8_t own[NFSV4CL_LOCKNAMELEN]; int error = 0; np = VTONFS(vp); switch (fl->l_whence) { case SEEK_SET: case SEEK_CUR: /* * Caller is responsible for adding any necessary offset * when SEEK_CUR is used. */ off = fl->l_start; break; case SEEK_END: off = np->n_size + fl->l_start; break; default: return (1); } if (fl->l_len != 0) { end = off + fl->l_len; if (end < off) return (1); } else { end = NFS64BITSSET; } error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp); if (error) return (1); nfscl_filllockowner(id, own, flags); NFSLOCKCLSTATE(); /* * First check the delegation locks. */ dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL) { LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) break; } if (lp != NULL) { LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) { if (lop->nfslo_first >= end) break; if (lop->nfslo_end <= off) continue; if (lop->nfslo_type == F_WRLCK) { nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); return (1); } } } } /* * Now, check state against the server. */ LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == np->n_fhp->nfh_len && !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) { LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) break; } if (lp != NULL) { LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) { if (lop->nfslo_first >= end) break; if (lop->nfslo_end <= off) continue; if (lop->nfslo_type == F_WRLCK) { nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); return (1); } } } } } } nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); return (0); } /* * Release a byte range lock owner structure. */ APPLESTATIC void nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete) { struct nfsclclient *clp; if (lp == NULL) return; NFSLOCKCLSTATE(); clp = lp->nfsl_open->nfso_own->nfsow_clp; if (error != 0 && candelete && (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0) nfscl_freelockowner(lp, 0); else nfscl_lockunlock(&lp->nfsl_rwlock); nfscl_clrelease(clp); NFSUNLOCKCLSTATE(); } /* * Free up an open structure and any associated byte range lock structures. */ APPLESTATIC void nfscl_freeopen(struct nfsclopen *op, int local) { LIST_REMOVE(op, nfso_list); nfscl_freealllocks(&op->nfso_lock, local); FREE((caddr_t)op, M_NFSCLOPEN); if (local) nfsstatsv1.cllocalopens--; else nfsstatsv1.clopens--; } /* * Free up all lock owners and associated locks. */ static void nfscl_freealllocks(struct nfscllockownerhead *lhp, int local) { struct nfscllockowner *lp, *nlp; LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) { if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED)) panic("nfscllckw"); nfscl_freelockowner(lp, local); } } /* * Called for an Open when NFSERR_EXPIRED is received from the server. * If there are no byte range locks nor a Share Deny lost, try to do a * fresh Open. Otherwise, free the open. */ static int nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op, struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p) { struct nfscllockowner *lp; struct nfscldeleg *dp; int mustdelete = 0, error; /* * Look for any byte range lock(s). */ LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (!LIST_EMPTY(&lp->nfsl_lock)) { mustdelete = 1; break; } } /* * If no byte range lock(s) nor a Share deny, try to re-open. */ if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) { newnfs_copycred(&op->nfso_cred, cred); dp = NULL; error = nfsrpc_reopen(nmp, op->nfso_fh, op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p); if (error) { mustdelete = 1; if (dp != NULL) { FREE((caddr_t)dp, M_NFSCLDELEG); dp = NULL; } } if (dp != NULL) nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh, op->nfso_fhlen, cred, p, &dp); } /* * If a byte range lock or Share deny or couldn't re-open, free it. */ if (mustdelete) nfscl_freeopen(op, 0); return (mustdelete); } /* * Free up an open owner structure. */ static void nfscl_freeopenowner(struct nfsclowner *owp, int local) { LIST_REMOVE(owp, nfsow_list); FREE((caddr_t)owp, M_NFSCLOWNER); if (local) nfsstatsv1.cllocalopenowners--; else nfsstatsv1.clopenowners--; } /* * Free up a byte range lock owner structure. */ APPLESTATIC void nfscl_freelockowner(struct nfscllockowner *lp, int local) { struct nfscllock *lop, *nlop; LIST_REMOVE(lp, nfsl_list); LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) { nfscl_freelock(lop, local); } FREE((caddr_t)lp, M_NFSCLLOCKOWNER); if (local) nfsstatsv1.cllocallockowners--; else nfsstatsv1.cllockowners--; } /* * Free up a byte range lock structure. */ APPLESTATIC void nfscl_freelock(struct nfscllock *lop, int local) { LIST_REMOVE(lop, nfslo_list); FREE((caddr_t)lop, M_NFSCLLOCK); if (local) nfsstatsv1.cllocallocks--; else nfsstatsv1.cllocks--; } /* * Clean out the state related to a delegation. */ static void nfscl_cleandeleg(struct nfscldeleg *dp) { struct nfsclowner *owp, *nowp; struct nfsclopen *op; LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) { op = LIST_FIRST(&owp->nfsow_open); if (op != NULL) { if (LIST_NEXT(op, nfso_list) != NULL) panic("nfscleandel"); nfscl_freeopen(op, 1); } nfscl_freeopenowner(owp, 1); } nfscl_freealllocks(&dp->nfsdl_lock, 1); } /* * Free a delegation. */ static void nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp) { TAILQ_REMOVE(hdp, dp, nfsdl_list); LIST_REMOVE(dp, nfsdl_hash); FREE((caddr_t)dp, M_NFSCLDELEG); nfsstatsv1.cldelegates--; nfscl_delegcnt--; } /* * Free up all state related to this client structure. */ static void nfscl_cleanclient(struct nfsclclient *clp) { struct nfsclowner *owp, *nowp; struct nfsclopen *op, *nop; /* Now, all the OpenOwners, etc. */ LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) { LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) { nfscl_freeopen(op, 0); } nfscl_freeopenowner(owp, 0); } } /* * Called when an NFSERR_EXPIRED is received from the server. */ static void nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p) { struct nfsclowner *owp, *nowp, *towp; struct nfsclopen *op, *nop, *top; struct nfscldeleg *dp, *ndp; int ret, printed = 0; /* * First, merge locally issued Opens into the list for the server. */ dp = TAILQ_FIRST(&clp->nfsc_deleg); while (dp != NULL) { ndp = TAILQ_NEXT(dp, nfsdl_list); owp = LIST_FIRST(&dp->nfsdl_owner); while (owp != NULL) { nowp = LIST_NEXT(owp, nfsow_list); op = LIST_FIRST(&owp->nfsow_open); if (op != NULL) { if (LIST_NEXT(op, nfso_list) != NULL) panic("nfsclexp"); LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) { if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) break; } if (towp != NULL) { /* Merge opens in */ LIST_FOREACH(top, &towp->nfsow_open, nfso_list) { if (top->nfso_fhlen == op->nfso_fhlen && !NFSBCMP(top->nfso_fh, op->nfso_fh, op->nfso_fhlen)) { top->nfso_mode |= op->nfso_mode; top->nfso_opencnt += op->nfso_opencnt; break; } } if (top == NULL) { /* Just add the open to the owner list */ LIST_REMOVE(op, nfso_list); op->nfso_own = towp; LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list); nfsstatsv1.cllocalopens--; nfsstatsv1.clopens++; } } else { /* Just add the openowner to the client list */ LIST_REMOVE(owp, nfsow_list); owp->nfsow_clp = clp; LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list); nfsstatsv1.cllocalopenowners--; nfsstatsv1.clopenowners++; nfsstatsv1.cllocalopens--; nfsstatsv1.clopens++; } } owp = nowp; } if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) { printed = 1; printf("nfsv4 expired locks lost\n"); } nfscl_cleandeleg(dp); nfscl_freedeleg(&clp->nfsc_deleg, dp); dp = ndp; } if (!TAILQ_EMPTY(&clp->nfsc_deleg)) panic("nfsclexp"); /* * Now, try and reopen against the server. */ LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) { owp->nfsow_seqid = 0; LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) { ret = nfscl_expireopen(clp, op, nmp, cred, p); if (ret && !printed) { printed = 1; printf("nfsv4 expired locks lost\n"); } } if (LIST_EMPTY(&owp->nfsow_open)) nfscl_freeopenowner(owp, 0); } } /* * This function must be called after the process represented by "own" has * exited. Must be called with CLSTATE lock held. */ static void nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own) { struct nfsclowner *owp, *nowp; struct nfscllockowner *lp, *nlp; struct nfscldeleg *dp; /* First, get rid of local locks on delegations. */ TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) { LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) { if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) { if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED)) panic("nfscllckw"); nfscl_freelockowner(lp, 1); } } } owp = LIST_FIRST(&clp->nfsc_owner); while (owp != NULL) { nowp = LIST_NEXT(owp, nfsow_list); if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN)) { /* * If there are children that haven't closed the * file descriptors yet, the opens will still be * here. For that case, let the renew thread clear * out the OpenOwner later. */ if (LIST_EMPTY(&owp->nfsow_open)) nfscl_freeopenowner(owp, 0); else owp->nfsow_defunct = 1; } owp = nowp; } } /* * Find open/lock owners for processes that have exited. */ static void nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp) { struct nfsclowner *owp, *nowp; struct nfsclopen *op; struct nfscllockowner *lp, *nlp; struct nfscldeleg *dp; NFSPROCLISTLOCK(); NFSLOCKCLSTATE(); LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) { if (LIST_EMPTY(&lp->nfsl_lock)) nfscl_emptylockowner(lp, lhp); } } if (nfscl_procdoesntexist(owp->nfsow_owner)) nfscl_cleanup_common(clp, owp->nfsow_owner); } /* * For the single open_owner case, these lock owners need to be * checked to see if they still exist separately. * This is because nfscl_procdoesntexist() never returns true for * the single open_owner so that the above doesn't ever call * nfscl_cleanup_common(). */ TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) { LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) { if (nfscl_procdoesntexist(lp->nfsl_owner)) nfscl_cleanup_common(clp, lp->nfsl_owner); } } NFSUNLOCKCLSTATE(); NFSPROCLISTUNLOCK(); } /* * Take the empty lock owner and move it to the local lhp list if the * associated process no longer exists. */ static void nfscl_emptylockowner(struct nfscllockowner *lp, struct nfscllockownerfhhead *lhp) { struct nfscllockownerfh *lfhp, *mylfhp; struct nfscllockowner *nlp; int fnd_it; /* If not a Posix lock owner, just return. */ if ((lp->nfsl_lockflags & F_POSIX) == 0) return; fnd_it = 0; mylfhp = NULL; /* * First, search to see if this lock owner is already in the list. * If it is, then the associated process no longer exists. */ SLIST_FOREACH(lfhp, lhp, nfslfh_list) { if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen && !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh, lfhp->nfslfh_len)) mylfhp = lfhp; LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list) if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner, NFSV4CL_LOCKNAMELEN)) fnd_it = 1; } /* If not found, check if process still exists. */ if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0) return; /* Move the lock owner over to the local list. */ if (mylfhp == NULL) { mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP, M_NOWAIT); if (mylfhp == NULL) return; mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen; NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh, mylfhp->nfslfh_len); LIST_INIT(&mylfhp->nfslfh_lock); SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list); } LIST_REMOVE(lp, nfsl_list); LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list); } static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */ /* * Called from nfs umount to free up the clientid. */ APPLESTATIC void nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p) { struct nfsclclient *clp; struct ucred *cred; int igotlock; /* * For the case that matters, this is the thread that set * MNTK_UNMOUNTF, so it will see it set. The code that follows is * done to ensure that any thread executing nfscl_getcl() after * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the * mutex for NFSLOCKCLSTATE(), so it is "m" for the following * explanation, courtesy of Alan Cox. * What follows is a snippet from Alan Cox's email at: * http://docs.FreeBSD.org/cgi/ * mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw * * 1. Set MNTK_UNMOUNTF * 2. Acquire a standard FreeBSD mutex "m". * 3. Update some data structures. * 4. Release mutex "m". * * Then, other threads that acquire "m" after step 4 has occurred will * see MNTK_UNMOUNTF as set. But, other threads that beat thread X to * step 2 may or may not see MNTK_UNMOUNTF as set. */ NFSLOCKCLSTATE(); if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { fake_global++; NFSUNLOCKCLSTATE(); NFSLOCKCLSTATE(); } clp = nmp->nm_clp; if (clp != NULL) { if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0) panic("nfscl umount"); /* * First, handshake with the nfscl renew thread, to terminate * it. */ clp->nfsc_flags |= NFSCLFLAGS_UMOUNT; while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD) (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfsclumnt", hz); /* * Now, get the exclusive lock on the client state, so * that no uses of the state are still in progress. */ do { igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL, NFSCLSTATEMUTEXPTR, NULL); } while (!igotlock); NFSUNLOCKCLSTATE(); /* * Free up all the state. It will expire on the server, but * maybe we should do a SetClientId/SetClientIdConfirm so * the server throws it away? */ LIST_REMOVE(clp, nfsc_list); nfscl_delegreturnall(clp, p); cred = newnfs_getcred(); if (NFSHASNFSV4N(nmp)) { (void)nfsrpc_destroysession(nmp, clp, cred, p); (void)nfsrpc_destroyclient(nmp, clp, cred, p); } else (void)nfsrpc_setclient(nmp, clp, 0, cred, p); nfscl_cleanclient(clp); nmp->nm_clp = NULL; NFSFREECRED(cred); free(clp, M_NFSCLCLIENT); } else NFSUNLOCKCLSTATE(); } /* * This function is called when a server replies with NFSERR_STALECLIENTID * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists, * doing Opens and Locks with reclaim. If these fail, it deletes the * corresponding state. */ static void nfscl_recover(struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p) { struct nfsclowner *owp, *nowp; struct nfsclopen *op, *nop; struct nfscllockowner *lp, *nlp; struct nfscllock *lop, *nlop; struct nfscldeleg *dp, *ndp, *tdp; struct nfsmount *nmp; struct ucred *tcred; struct nfsclopenhead extra_open; struct nfscldeleghead extra_deleg; struct nfsreq *rep; u_int64_t len; u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode; int i, igotlock = 0, error, trycnt, firstlock; struct nfscllayout *lyp, *nlyp; /* * First, lock the client structure, so everyone else will * block when trying to use state. */ NFSLOCKCLSTATE(); clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG; do { igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL, NFSCLSTATEMUTEXPTR, NULL); } while (!igotlock); NFSUNLOCKCLSTATE(); nmp = clp->nfsc_nmp; if (nmp == NULL) panic("nfscl recover"); /* * For now, just get rid of all layouts. There may be a need * to do LayoutCommit Ops with reclaim == true later. */ TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) nfscl_freelayout(lyp); TAILQ_INIT(&clp->nfsc_layout); for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++) LIST_INIT(&clp->nfsc_layouthash[i]); trycnt = 5; do { error = nfsrpc_setclient(nmp, clp, 1, cred, p); } while ((error == NFSERR_STALECLIENTID || error == NFSERR_BADSESSION || error == NFSERR_STALEDONTRECOVER) && --trycnt > 0); if (error) { NFSLOCKCLSTATE(); clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER | NFSCLFLAGS_RECVRINPROG); wakeup(&clp->nfsc_flags); nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); return; } clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID; clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER; /* * Mark requests already queued on the server, so that they don't * initiate another recovery cycle. Any requests already in the * queue that handle state information will have the old stale * clientid/stateid and will get a NFSERR_STALESTATEID, * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server. * This will be translated to NFSERR_STALEDONTRECOVER when * R_DONTRECOVER is set. */ NFSLOCKREQ(); TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) { if (rep->r_nmp == nmp) rep->r_flags |= R_DONTRECOVER; } NFSUNLOCKREQ(); /* * Now, mark all delegations "need reclaim". */ TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM; TAILQ_INIT(&extra_deleg); LIST_INIT(&extra_open); /* * Now traverse the state lists, doing Open and Lock Reclaims. */ tcred = newnfs_getcred(); owp = LIST_FIRST(&clp->nfsc_owner); while (owp != NULL) { nowp = LIST_NEXT(owp, nfsow_list); owp->nfsow_seqid = 0; op = LIST_FIRST(&owp->nfsow_open); while (op != NULL) { nop = LIST_NEXT(op, nfso_list); if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) { /* Search for a delegation to reclaim with the open */ TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) { if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) continue; if ((dp->nfsdl_flags & NFSCLDL_WRITE)) { mode = NFSV4OPEN_ACCESSWRITE; delegtype = NFSV4OPEN_DELEGATEWRITE; } else { mode = NFSV4OPEN_ACCESSREAD; delegtype = NFSV4OPEN_DELEGATEREAD; } if ((op->nfso_mode & mode) == mode && op->nfso_fhlen == dp->nfsdl_fhlen && !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen)) break; } ndp = dp; if (dp == NULL) delegtype = NFSV4OPEN_DELEGATENONE; newnfs_copycred(&op->nfso_cred, tcred); error = nfscl_tryopen(nmp, NULL, op->nfso_fh, op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen, op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype, tcred, p); if (!error) { /* Handle any replied delegation */ if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE) || NFSMNT_RDONLY(nmp->nm_mountp))) { if ((ndp->nfsdl_flags & NFSCLDL_WRITE)) mode = NFSV4OPEN_ACCESSWRITE; else mode = NFSV4OPEN_ACCESSREAD; TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) { if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) continue; if ((op->nfso_mode & mode) == mode && op->nfso_fhlen == dp->nfsdl_fhlen && !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen)) { dp->nfsdl_stateid = ndp->nfsdl_stateid; dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit; dp->nfsdl_ace = ndp->nfsdl_ace; dp->nfsdl_change = ndp->nfsdl_change; dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM; if ((ndp->nfsdl_flags & NFSCLDL_RECALL)) dp->nfsdl_flags |= NFSCLDL_RECALL; FREE((caddr_t)ndp, M_NFSCLDELEG); ndp = NULL; break; } } } if (ndp != NULL) TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list); /* and reclaim all byte range locks */ lp = LIST_FIRST(&op->nfso_lock); while (lp != NULL) { nlp = LIST_NEXT(lp, nfsl_list); lp->nfsl_seqid = 0; firstlock = 1; lop = LIST_FIRST(&lp->nfsl_lock); while (lop != NULL) { nlop = LIST_NEXT(lop, nfslo_list); if (lop->nfslo_end == NFS64BITSSET) len = NFS64BITSSET; else len = lop->nfslo_end - lop->nfslo_first; error = nfscl_trylock(nmp, NULL, op->nfso_fh, op->nfso_fhlen, lp, firstlock, 1, lop->nfslo_first, len, lop->nfslo_type, tcred, p); if (error != 0) nfscl_freelock(lop, 0); else firstlock = 0; lop = nlop; } /* If no locks, but a lockowner, just delete it. */ if (LIST_EMPTY(&lp->nfsl_lock)) nfscl_freelockowner(lp, 0); lp = nlp; } } } if (error != 0 && error != NFSERR_BADSESSION) nfscl_freeopen(op, 0); op = nop; } owp = nowp; } /* * Now, try and get any delegations not yet reclaimed by cobbling * to-gether an appropriate open. */ nowp = NULL; dp = TAILQ_FIRST(&clp->nfsc_deleg); while (dp != NULL) { ndp = TAILQ_NEXT(dp, nfsdl_list); if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) { if (nowp == NULL) { MALLOC(nowp, struct nfsclowner *, sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK); /* * Name must be as long an largest possible * NFSV4CL_LOCKNAMELEN. 12 for now. */ NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN); LIST_INIT(&nowp->nfsow_open); nowp->nfsow_clp = clp; nowp->nfsow_seqid = 0; nowp->nfsow_defunct = 0; nfscl_lockinit(&nowp->nfsow_rwlock); } nop = NULL; if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) { MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) + dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK); nop->nfso_own = nowp; if ((dp->nfsdl_flags & NFSCLDL_WRITE)) { nop->nfso_mode = NFSV4OPEN_ACCESSWRITE; delegtype = NFSV4OPEN_DELEGATEWRITE; } else { nop->nfso_mode = NFSV4OPEN_ACCESSREAD; delegtype = NFSV4OPEN_DELEGATEREAD; } nop->nfso_opencnt = 0; nop->nfso_posixlock = 1; nop->nfso_fhlen = dp->nfsdl_fhlen; NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen); LIST_INIT(&nop->nfso_lock); nop->nfso_stateid.seqid = 0; nop->nfso_stateid.other[0] = 0; nop->nfso_stateid.other[1] = 0; nop->nfso_stateid.other[2] = 0; newnfs_copycred(&dp->nfsdl_cred, tcred); newnfs_copyincred(tcred, &nop->nfso_cred); tdp = NULL; error = nfscl_tryopen(nmp, NULL, nop->nfso_fh, nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen, nop->nfso_mode, nop, NULL, 0, &tdp, 1, delegtype, tcred, p); if (tdp != NULL) { if ((tdp->nfsdl_flags & NFSCLDL_WRITE)) mode = NFSV4OPEN_ACCESSWRITE; else mode = NFSV4OPEN_ACCESSREAD; if ((nop->nfso_mode & mode) == mode && nop->nfso_fhlen == tdp->nfsdl_fhlen && !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh, nop->nfso_fhlen)) { dp->nfsdl_stateid = tdp->nfsdl_stateid; dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit; dp->nfsdl_ace = tdp->nfsdl_ace; dp->nfsdl_change = tdp->nfsdl_change; dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM; if ((tdp->nfsdl_flags & NFSCLDL_RECALL)) dp->nfsdl_flags |= NFSCLDL_RECALL; FREE((caddr_t)tdp, M_NFSCLDELEG); } else { TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list); } } } if (error) { if (nop != NULL) FREE((caddr_t)nop, M_NFSCLOPEN); /* * Couldn't reclaim it, so throw the state * away. Ouch!! */ nfscl_cleandeleg(dp); nfscl_freedeleg(&clp->nfsc_deleg, dp); } else { LIST_INSERT_HEAD(&extra_open, nop, nfso_list); } } dp = ndp; } /* * Now, get rid of extra Opens and Delegations. */ LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) { do { newnfs_copycred(&op->nfso_cred, tcred); error = nfscl_tryclose(op, tcred, nmp, p); if (error == NFSERR_GRACE) (void) nfs_catnap(PZERO, error, "nfsexcls"); } while (error == NFSERR_GRACE); LIST_REMOVE(op, nfso_list); FREE((caddr_t)op, M_NFSCLOPEN); } if (nowp != NULL) FREE((caddr_t)nowp, M_NFSCLOWNER); TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) { do { newnfs_copycred(&dp->nfsdl_cred, tcred); error = nfscl_trydelegreturn(dp, tcred, nmp, p); if (error == NFSERR_GRACE) (void) nfs_catnap(PZERO, error, "nfsexdlg"); } while (error == NFSERR_GRACE); TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list); FREE((caddr_t)dp, M_NFSCLDELEG); } /* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */ if (NFSHASNFSV4N(nmp)) (void)nfsrpc_reclaimcomplete(nmp, cred, p); NFSLOCKCLSTATE(); clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG; wakeup(&clp->nfsc_flags); nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); NFSFREECRED(tcred); } /* * This function is called when a server replies with NFSERR_EXPIRED. * It deletes all state for the client and does a fresh SetClientId/confirm. * XXX Someday it should post a signal to the process(es) that hold the * state, so they know that lock state has been lost. */ APPLESTATIC int nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p) { struct nfsmount *nmp; struct ucred *cred; int igotlock = 0, error, trycnt; /* * If the clientid has gone away or a new SetClientid has already * been done, just return ok. */ if (clp == NULL || clidrev != clp->nfsc_clientidrev) return (0); /* * First, lock the client structure, so everyone else will * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so * that only one thread does the work. */ NFSLOCKCLSTATE(); clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT; do { igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL, NFSCLSTATEMUTEXPTR, NULL); } while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT)); if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) { if (igotlock) nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); return (0); } clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG; NFSUNLOCKCLSTATE(); nmp = clp->nfsc_nmp; if (nmp == NULL) panic("nfscl expired"); cred = newnfs_getcred(); trycnt = 5; do { error = nfsrpc_setclient(nmp, clp, 0, cred, p); } while ((error == NFSERR_STALECLIENTID || error == NFSERR_BADSESSION || error == NFSERR_STALEDONTRECOVER) && --trycnt > 0); if (error) { NFSLOCKCLSTATE(); clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER; } else { /* * Expire the state for the client. */ nfscl_expireclient(clp, nmp, cred, p); NFSLOCKCLSTATE(); clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID; clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER; } clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG); wakeup(&clp->nfsc_flags); nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); NFSFREECRED(cred); return (error); } /* * This function inserts a lock in the list after insert_lop. */ static void nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop, struct nfscllock *insert_lop, int local) { if ((struct nfscllockowner *)insert_lop == lp) LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list); else LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list); if (local) nfsstatsv1.cllocallocks++; else nfsstatsv1.cllocks++; } /* * This function updates the locking for a lock owner and given file. It * maintains a list of lock ranges ordered on increasing file offset that * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style). * It always adds new_lop to the list and sometimes uses the one pointed * at by other_lopp. * Returns 1 if the locks were modified, 0 otherwise. */ static int nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp, struct nfscllock **other_lopp, int local) { struct nfscllock *new_lop = *new_lopp; struct nfscllock *lop, *tlop, *ilop; struct nfscllock *other_lop; int unlock = 0, modified = 0; u_int64_t tmp; /* * Work down the list until the lock is merged. */ if (new_lop->nfslo_type == F_UNLCK) unlock = 1; ilop = (struct nfscllock *)lp; lop = LIST_FIRST(&lp->nfsl_lock); while (lop != NULL) { /* * Only check locks for this file that aren't before the start of * new lock's range. */ if (lop->nfslo_end >= new_lop->nfslo_first) { if (new_lop->nfslo_end < lop->nfslo_first) { /* * If the new lock ends before the start of the * current lock's range, no merge, just insert * the new lock. */ break; } if (new_lop->nfslo_type == lop->nfslo_type || (new_lop->nfslo_first <= lop->nfslo_first && new_lop->nfslo_end >= lop->nfslo_end)) { /* * This lock can be absorbed by the new lock/unlock. * This happens when it covers the entire range * of the old lock or is contiguous * with the old lock and is of the same type or an * unlock. */ if (new_lop->nfslo_type != lop->nfslo_type || new_lop->nfslo_first != lop->nfslo_first || new_lop->nfslo_end != lop->nfslo_end) modified = 1; if (lop->nfslo_first < new_lop->nfslo_first) new_lop->nfslo_first = lop->nfslo_first; if (lop->nfslo_end > new_lop->nfslo_end) new_lop->nfslo_end = lop->nfslo_end; tlop = lop; lop = LIST_NEXT(lop, nfslo_list); nfscl_freelock(tlop, local); continue; } /* * All these cases are for contiguous locks that are not the * same type, so they can't be merged. */ if (new_lop->nfslo_first <= lop->nfslo_first) { /* * This case is where the new lock overlaps with the * first part of the old lock. Move the start of the * old lock to just past the end of the new lock. The * new lock will be inserted in front of the old, since * ilop hasn't been updated. (We are done now.) */ if (lop->nfslo_first != new_lop->nfslo_end) { lop->nfslo_first = new_lop->nfslo_end; modified = 1; } break; } if (new_lop->nfslo_end >= lop->nfslo_end) { /* * This case is where the new lock overlaps with the * end of the old lock's range. Move the old lock's * end to just before the new lock's first and insert * the new lock after the old lock. * Might not be done yet, since the new lock could * overlap further locks with higher ranges. */ if (lop->nfslo_end != new_lop->nfslo_first) { lop->nfslo_end = new_lop->nfslo_first; modified = 1; } ilop = lop; lop = LIST_NEXT(lop, nfslo_list); continue; } /* * The final case is where the new lock's range is in the * middle of the current lock's and splits the current lock * up. Use *other_lopp to handle the second part of the * split old lock range. (We are done now.) * For unlock, we use new_lop as other_lop and tmp, since * other_lop and new_lop are the same for this case. * We noted the unlock case above, so we don't need * new_lop->nfslo_type any longer. */ tmp = new_lop->nfslo_first; if (unlock) { other_lop = new_lop; *new_lopp = NULL; } else { other_lop = *other_lopp; *other_lopp = NULL; } other_lop->nfslo_first = new_lop->nfslo_end; other_lop->nfslo_end = lop->nfslo_end; other_lop->nfslo_type = lop->nfslo_type; lop->nfslo_end = tmp; nfscl_insertlock(lp, other_lop, lop, local); ilop = lop; modified = 1; break; } ilop = lop; lop = LIST_NEXT(lop, nfslo_list); if (lop == NULL) break; } /* * Insert the new lock in the list at the appropriate place. */ if (!unlock) { nfscl_insertlock(lp, new_lop, ilop, local); *new_lopp = NULL; modified = 1; } return (modified); } /* * This function must be run as a kernel thread. * It does Renew Ops and recovery, when required. */ APPLESTATIC void nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p) { struct nfsclowner *owp, *nowp; struct nfsclopen *op; struct nfscllockowner *lp, *nlp; struct nfscldeleghead dh; struct nfscldeleg *dp, *ndp; struct ucred *cred; u_int32_t clidrev; int error, cbpathdown, islept, igotlock, ret, clearok; uint32_t recover_done_time = 0; time_t mytime; static time_t prevsec = 0; struct nfscllockownerfh *lfhp, *nlfhp; struct nfscllockownerfhhead lfh; struct nfscllayout *lyp, *nlyp; struct nfscldevinfo *dip, *ndip; struct nfscllayouthead rlh; struct nfsclrecalllayout *recallp; struct nfsclds *dsp; cred = newnfs_getcred(); NFSLOCKCLSTATE(); clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD; NFSUNLOCKCLSTATE(); for(;;) { newnfs_setroot(cred); cbpathdown = 0; if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) { /* * Only allow one recover within 1/2 of the lease * duration (nfsc_renew). */ if (recover_done_time < NFSD_MONOSEC) { recover_done_time = NFSD_MONOSEC + clp->nfsc_renew; NFSCL_DEBUG(1, "Doing recovery..\n"); nfscl_recover(clp, cred, p); } else { NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n", recover_done_time, (intmax_t)NFSD_MONOSEC); NFSLOCKCLSTATE(); clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER; NFSUNLOCKCLSTATE(); } } if (clp->nfsc_expire <= NFSD_MONOSEC && (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) { clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew; clidrev = clp->nfsc_clientidrev; error = nfsrpc_renew(clp, NULL, cred, p); if (error == NFSERR_CBPATHDOWN) cbpathdown = 1; else if (error == NFSERR_STALECLIENTID || error == NFSERR_BADSESSION) { NFSLOCKCLSTATE(); clp->nfsc_flags |= NFSCLFLAGS_RECOVER; NFSUNLOCKCLSTATE(); } else if (error == NFSERR_EXPIRED) (void) nfscl_hasexpired(clp, clidrev, p); } checkdsrenew: if (NFSHASNFSV4N(clp->nfsc_nmp)) { /* Do renews for any DS sessions. */ NFSLOCKMNT(clp->nfsc_nmp); /* Skip first entry, since the MDS is handled above. */ dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess); if (dsp != NULL) dsp = TAILQ_NEXT(dsp, nfsclds_list); while (dsp != NULL) { if (dsp->nfsclds_expire <= NFSD_MONOSEC && dsp->nfsclds_sess.nfsess_defunct == 0) { dsp->nfsclds_expire = NFSD_MONOSEC + clp->nfsc_renew; NFSUNLOCKMNT(clp->nfsc_nmp); (void)nfsrpc_renew(clp, dsp, cred, p); goto checkdsrenew; } dsp = TAILQ_NEXT(dsp, nfsclds_list); } NFSUNLOCKMNT(clp->nfsc_nmp); } TAILQ_INIT(&dh); NFSLOCKCLSTATE(); if (cbpathdown) /* It's a Total Recall! */ nfscl_totalrecall(clp); /* * Now, handle defunct owners. */ LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) { if (LIST_EMPTY(&owp->nfsow_open)) { if (owp->nfsow_defunct != 0) nfscl_freeopenowner(owp, 0); } } /* * Do the recall on any delegations. To avoid trouble, always * come back up here after having slept. */ igotlock = 0; tryagain: dp = TAILQ_FIRST(&clp->nfsc_deleg); while (dp != NULL) { ndp = TAILQ_NEXT(dp, nfsdl_list); if ((dp->nfsdl_flags & NFSCLDL_RECALL)) { /* * Wait for outstanding I/O ops to be done. */ if (dp->nfsdl_rwlock.nfslock_usecnt > 0) { if (igotlock) { nfsv4_unlock(&clp->nfsc_lock, 0); igotlock = 0; } dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED; (void) nfsmsleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL); goto tryagain; } while (!igotlock) { igotlock = nfsv4_lock(&clp->nfsc_lock, 1, &islept, NFSCLSTATEMUTEXPTR, NULL); if (islept) goto tryagain; } NFSUNLOCKCLSTATE(); newnfs_copycred(&dp->nfsdl_cred, cred); ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp, NULL, cred, p, 1); if (!ret) { nfscl_cleandeleg(dp); TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list); LIST_REMOVE(dp, nfsdl_hash); TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list); nfscl_delegcnt--; nfsstatsv1.cldelegates--; } NFSLOCKCLSTATE(); } dp = ndp; } /* * Clear out old delegations, if we are above the high water * mark. Only clear out ones with no state related to them. * The tailq list is in LRU order. */ dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead); while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) { ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list); if (dp->nfsdl_rwlock.nfslock_usecnt == 0 && dp->nfsdl_rwlock.nfslock_lock == 0 && dp->nfsdl_timestamp < NFSD_MONOSEC && (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED | NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) { clearok = 1; LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) { op = LIST_FIRST(&owp->nfsow_open); if (op != NULL) { clearok = 0; break; } } if (clearok) { LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { if (!LIST_EMPTY(&lp->nfsl_lock)) { clearok = 0; break; } } } if (clearok) { TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list); LIST_REMOVE(dp, nfsdl_hash); TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list); nfscl_delegcnt--; nfsstatsv1.cldelegates--; } } dp = ndp; } if (igotlock) nfsv4_unlock(&clp->nfsc_lock, 0); /* * Do the recall on any layouts. To avoid trouble, always * come back up here after having slept. */ TAILQ_INIT(&rlh); tryagain2: TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) { if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) { /* * Wait for outstanding I/O ops to be done. */ if (lyp->nfsly_lock.nfslock_usecnt > 0 || (lyp->nfsly_lock.nfslock_lock & NFSV4LOCK_LOCK) != 0) { lyp->nfsly_lock.nfslock_lock |= NFSV4LOCK_WANTED; (void)nfsmsleep(&lyp->nfsly_lock, NFSCLSTATEMUTEXPTR, PZERO, "nfslyp", NULL); goto tryagain2; } /* Move the layout to the recall list. */ TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list); LIST_REMOVE(lyp, nfsly_hash); TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list); /* Handle any layout commits. */ if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) && (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) { lyp->nfsly_flags &= ~NFSLY_WRITTEN; NFSUNLOCKCLSTATE(); NFSCL_DEBUG(3, "do layoutcommit\n"); nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, cred, p); NFSLOCKCLSTATE(); goto tryagain2; } } } /* Now, look for stale layouts. */ lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead); while (lyp != NULL) { nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list); if (lyp->nfsly_timestamp < NFSD_MONOSEC && (lyp->nfsly_flags & NFSLY_RECALL) == 0 && lyp->nfsly_lock.nfslock_usecnt == 0 && lyp->nfsly_lock.nfslock_lock == 0) { NFSCL_DEBUG(4, "ret stale lay=%d\n", nfscl_layoutcnt); recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_NOWAIT); if (recallp == NULL) break; (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX, lyp->nfsly_stateid.seqid, recallp); } lyp = nlyp; } /* * Free up any unreferenced device info structures. */ LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) { if (dip->nfsdi_layoutrefs == 0 && dip->nfsdi_refcnt == 0) { NFSCL_DEBUG(4, "freeing devinfo\n"); LIST_REMOVE(dip, nfsdi_list); nfscl_freedevinfo(dip); } } NFSUNLOCKCLSTATE(); /* Do layout return(s), as required. */ TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) { TAILQ_REMOVE(&rlh, lyp, nfsly_list); NFSCL_DEBUG(4, "ret layout\n"); nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p); nfscl_freelayout(lyp); } /* * Delegreturn any delegations cleaned out or recalled. */ TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) { newnfs_copycred(&dp->nfsdl_cred, cred); (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p); TAILQ_REMOVE(&dh, dp, nfsdl_list); FREE((caddr_t)dp, M_NFSCLDELEG); } SLIST_INIT(&lfh); /* * Call nfscl_cleanupkext() once per second to check for * open/lock owners where the process has exited. */ mytime = NFSD_MONOSEC; if (prevsec != mytime) { prevsec = mytime; nfscl_cleanupkext(clp, &lfh); } /* * Do a ReleaseLockOwner for all lock owners where the * associated process no longer exists, as found by * nfscl_cleanupkext(). */ newnfs_setroot(cred); SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) { LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list, nlp) { (void)nfsrpc_rellockown(clp->nfsc_nmp, lp, lfhp->nfslfh_fh, lfhp->nfslfh_len, cred, p); nfscl_freelockowner(lp, 0); } free(lfhp, M_TEMP); } SLIST_INIT(&lfh); NFSLOCKCLSTATE(); if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0) (void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl", hz); if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) { clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD; NFSUNLOCKCLSTATE(); NFSFREECRED(cred); wakeup((caddr_t)clp); return; } NFSUNLOCKCLSTATE(); } } /* * Initiate state recovery. Called when NFSERR_STALECLIENTID, * NFSERR_STALESTATEID or NFSERR_BADSESSION is received. */ APPLESTATIC void nfscl_initiate_recovery(struct nfsclclient *clp) { if (clp == NULL) return; NFSLOCKCLSTATE(); clp->nfsc_flags |= NFSCLFLAGS_RECOVER; NFSUNLOCKCLSTATE(); wakeup((caddr_t)clp); } /* * Dump out the state stuff for debugging. */ APPLESTATIC void nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens, int lockowner, int locks) { struct nfsclclient *clp; struct nfsclowner *owp; struct nfsclopen *op; struct nfscllockowner *lp; struct nfscllock *lop; struct nfscldeleg *dp; clp = nmp->nm_clp; if (clp == NULL) { printf("nfscl dumpstate NULL clp\n"); return; } NFSLOCKCLSTATE(); TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) { LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) { if (openowner && !LIST_EMPTY(&owp->nfsow_open)) printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n", owp->nfsow_owner[0], owp->nfsow_owner[1], owp->nfsow_owner[2], owp->nfsow_owner[3], owp->nfsow_seqid); LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (opens) printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n", op->nfso_stateid.other[0], op->nfso_stateid.other[1], op->nfso_stateid.other[2], op->nfso_opencnt, op->nfso_fh[12]); LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (lockowner) printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n", lp->nfsl_owner[0], lp->nfsl_owner[1], lp->nfsl_owner[2], lp->nfsl_owner[3], lp->nfsl_seqid, lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1], lp->nfsl_stateid.other[2]); LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) { if (locks) #ifdef __FreeBSD__ printf("lck typ=%d fst=%ju end=%ju\n", lop->nfslo_type, (intmax_t)lop->nfslo_first, (intmax_t)lop->nfslo_end); #else printf("lck typ=%d fst=%qd end=%qd\n", lop->nfslo_type, lop->nfslo_first, lop->nfslo_end); #endif } } } } } LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { if (openowner && !LIST_EMPTY(&owp->nfsow_open)) printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n", owp->nfsow_owner[0], owp->nfsow_owner[1], owp->nfsow_owner[2], owp->nfsow_owner[3], owp->nfsow_seqid); LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (opens) printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n", op->nfso_stateid.other[0], op->nfso_stateid.other[1], op->nfso_stateid.other[2], op->nfso_opencnt, op->nfso_fh[12]); LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { if (lockowner) printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n", lp->nfsl_owner[0], lp->nfsl_owner[1], lp->nfsl_owner[2], lp->nfsl_owner[3], lp->nfsl_seqid, lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1], lp->nfsl_stateid.other[2]); LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) { if (locks) #ifdef __FreeBSD__ printf("lck typ=%d fst=%ju end=%ju\n", lop->nfslo_type, (intmax_t)lop->nfslo_first, (intmax_t)lop->nfslo_end); #else printf("lck typ=%d fst=%qd end=%qd\n", lop->nfslo_type, lop->nfslo_first, lop->nfslo_end); #endif } } } } NFSUNLOCKCLSTATE(); } /* * Check for duplicate open owners and opens. * (Only used as a diagnostic aid.) */ APPLESTATIC void nfscl_dupopen(vnode_t vp, int dupopens) { struct nfsclclient *clp; struct nfsclowner *owp, *owp2; struct nfsclopen *op, *op2; struct nfsfh *nfhp; clp = VFSTONFS(vnode_mount(vp))->nm_clp; if (clp == NULL) { printf("nfscl dupopen NULL clp\n"); return; } nfhp = VTONFS(vp)->n_fhp; NFSLOCKCLSTATE(); /* * First, search for duplicate owners. * These should never happen! */ LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { if (owp != owp2 && !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner, NFSV4CL_LOCKNAMELEN)) { NFSUNLOCKCLSTATE(); printf("DUP OWNER\n"); nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0); return; } } } /* * Now, search for duplicate stateids. * These shouldn't happen, either. */ LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) { LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op != op2 && (op->nfso_stateid.other[0] != 0 || op->nfso_stateid.other[1] != 0 || op->nfso_stateid.other[2] != 0) && op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] && op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] && op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) { NFSUNLOCKCLSTATE(); printf("DUP STATEID\n"); nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0); return; } } } } } /* * Now search for duplicate opens. * Duplicate opens for the same owner * should never occur. Other duplicates are * possible and are checked for if "dupopens" * is true. */ LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) { if (nfhp->nfh_len == op2->nfso_fhlen && !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) { LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op != op2 && nfhp->nfh_len == op->nfso_fhlen && !NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) && (!NFSBCMP(op->nfso_own->nfsow_owner, op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) || dupopens)) { if (!NFSBCMP(op->nfso_own->nfsow_owner, op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) { NFSUNLOCKCLSTATE(); printf("BADDUP OPEN\n"); } else { NFSUNLOCKCLSTATE(); printf("DUP OPEN\n"); } nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0); return; } } } } } } NFSUNLOCKCLSTATE(); } /* * During close, find an open that needs to be dereferenced and * dereference it. If there are no more opens for this file, * log a message to that effect. * Opens aren't actually Close'd until VOP_INACTIVE() is performed * on the file's vnode. * This is the safe way, since it is difficult to identify * which open the close is for and I/O can be performed after the * close(2) system call when a file is mmap'd. * If it returns 0 for success, there will be a referenced * clp returned via clpp. */ APPLESTATIC int nfscl_getclose(vnode_t vp, struct nfsclclient **clpp) { struct nfsclclient *clp; struct nfsclowner *owp; struct nfsclopen *op; struct nfscldeleg *dp; struct nfsfh *nfhp; int error, notdecr; error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp); if (error) return (error); *clpp = clp; nfhp = VTONFS(vp)->n_fhp; notdecr = 1; NFSLOCKCLSTATE(); /* * First, look for one under a delegation that was locally issued * and just decrement the opencnt for it. Since all my Opens against * the server are DENY_NONE, I don't see a problem with hanging * onto them. (It is much easier to use one of the extant Opens * that I already have on the server when a Delegation is recalled * than to do fresh Opens.) Someday, I might need to rethink this, but. */ dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len); if (dp != NULL) { LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) { op = LIST_FIRST(&owp->nfsow_open); if (op != NULL) { /* * Since a delegation is for a file, there * should never be more than one open for * each openowner. */ if (LIST_NEXT(op, nfso_list) != NULL) panic("nfscdeleg opens"); if (notdecr && op->nfso_opencnt > 0) { notdecr = 0; op->nfso_opencnt--; break; } } } } /* Now process the opens against the server. */ LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == nfhp->nfh_len && !NFSBCMP(op->nfso_fh, nfhp->nfh_fh, nfhp->nfh_len)) { /* Found an open, decrement cnt if possible */ if (notdecr && op->nfso_opencnt > 0) { notdecr = 0; op->nfso_opencnt--; } /* * There are more opens, so just return. */ if (op->nfso_opencnt > 0) { NFSUNLOCKCLSTATE(); return (0); } } } } NFSUNLOCKCLSTATE(); if (notdecr) printf("nfscl: never fnd open\n"); return (0); } APPLESTATIC int nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p) { struct nfsclclient *clp; struct nfsclowner *owp, *nowp; struct nfsclopen *op; struct nfscldeleg *dp; struct nfsfh *nfhp; struct nfsclrecalllayout *recallp; int error; error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp); if (error) return (error); *clpp = clp; nfhp = VTONFS(vp)->n_fhp; recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK); NFSLOCKCLSTATE(); /* * First get rid of the local Open structures, which should be no * longer in use. */ dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len); if (dp != NULL) { LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) { op = LIST_FIRST(&owp->nfsow_open); if (op != NULL) { KASSERT((op->nfso_opencnt == 0), ("nfscl: bad open cnt on deleg")); nfscl_freeopen(op, 1); } nfscl_freeopenowner(owp, 1); } } /* Return any layouts marked return on close. */ nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp); /* Now process the opens against the server. */ lookformore: LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { op = LIST_FIRST(&owp->nfsow_open); while (op != NULL) { if (op->nfso_fhlen == nfhp->nfh_len && !NFSBCMP(op->nfso_fh, nfhp->nfh_fh, nfhp->nfh_len)) { /* Found an open, close it. */ KASSERT((op->nfso_opencnt == 0), ("nfscl: bad open cnt on server")); NFSUNLOCKCLSTATE(); nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op, p); NFSLOCKCLSTATE(); goto lookformore; } op = LIST_NEXT(op, nfso_list); } } NFSUNLOCKCLSTATE(); /* * recallp has been set NULL by nfscl_retoncloselayout() if it was * used by the function, but calling free() with a NULL pointer is ok. */ free(recallp, M_NFSLAYRECALL); return (0); } /* * Return all delegations on this client. * (Must be called with client sleep lock.) */ static void nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p) { struct nfscldeleg *dp, *ndp; struct ucred *cred; cred = newnfs_getcred(); TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) { nfscl_cleandeleg(dp); (void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p); nfscl_freedeleg(&clp->nfsc_deleg, dp); } NFSFREECRED(cred); } /* * Do a callback RPC. */ APPLESTATIC void nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p) { int clist, gotseq_ok, i, j, k, op, rcalls; u_int32_t *tl; struct nfsclclient *clp; struct nfscldeleg *dp = NULL; int numops, taglen = -1, error = 0, trunc; u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident; u_char tag[NFSV4_SMALLSTR + 1], *tagstr; vnode_t vp = NULL; struct nfsnode *np; struct vattr va; struct nfsfh *nfhp; mount_t mp; nfsattrbit_t attrbits, rattrbits; nfsv4stateid_t stateid; uint32_t seqid, slotid = 0, highslot, cachethis; uint8_t sessionid[NFSX_V4SESSIONID]; struct mbuf *rep; struct nfscllayout *lyp; uint64_t filesid[2], len, off; int changed, gotone, laytype, recalltype; uint32_t iomode; struct nfsclrecalllayout *recallp = NULL; struct nfsclsession *tsep; gotseq_ok = 0; nfsrvd_rephead(nd); NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); taglen = fxdr_unsigned(int, *tl); if (taglen < 0) { error = EBADRPC; goto nfsmout; } if (taglen <= NFSV4_SMALLSTR) tagstr = tag; else tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK); error = nfsrv_mtostr(nd, tagstr, taglen); if (error) { if (taglen > NFSV4_SMALLSTR) free(tagstr, M_TEMP); taglen = -1; goto nfsmout; } (void) nfsm_strtom(nd, tag, taglen); if (taglen > NFSV4_SMALLSTR) { free(tagstr, M_TEMP); } NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED); NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); minorvers = fxdr_unsigned(u_int32_t, *tl++); if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION) nd->nd_repstat = NFSERR_MINORVERMISMATCH; cbident = fxdr_unsigned(u_int32_t, *tl++); if (nd->nd_repstat) numops = 0; else numops = fxdr_unsigned(int, *tl); /* * Loop around doing the sub ops. */ for (i = 0; i < numops; i++) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED); *repp++ = *tl; op = fxdr_unsigned(int, *tl); if (op < NFSV4OP_CBGETATTR || (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) || (op > NFSV4OP_CBNOTIFYDEVID && minorvers == NFSV41_MINORVERSION)) { nd->nd_repstat = NFSERR_OPILLEGAL; *repp = nfscl_errmap(nd, minorvers); retops++; break; } nd->nd_procnum = op; if (op < NFSV41_CBNOPS) nfsstatsv1.cbrpccnt[nd->nd_procnum]++; switch (op) { case NFSV4OP_CBGETATTR: NFSCL_DEBUG(4, "cbgetattr\n"); mp = NULL; vp = NULL; error = nfsm_getfh(nd, &nfhp); if (!error) error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error == 0 && i == 0 && minorvers != NFSV4_MINORVERSION) error = NFSERR_OPNOTINSESS; if (!error) { mp = nfscl_getmnt(minorvers, sessionid, cbident, &clp); if (mp == NULL) error = NFSERR_SERVERFAULT; } if (!error) { error = nfscl_ngetreopen(mp, nfhp->nfh_fh, nfhp->nfh_len, p, &np); if (!error) vp = NFSTOV(np); } if (!error) { NFSZERO_ATTRBIT(&rattrbits); NFSLOCKCLSTATE(); dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len); if (dp != NULL) { if (NFSISSET_ATTRBIT(&attrbits, NFSATTRBIT_SIZE)) { if (vp != NULL) va.va_size = np->n_size; else va.va_size = dp->nfsdl_size; NFSSETBIT_ATTRBIT(&rattrbits, NFSATTRBIT_SIZE); } if (NFSISSET_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE)) { va.va_filerev = dp->nfsdl_change; if (vp == NULL || (np->n_flag & NDELEGMOD)) va.va_filerev++; NFSSETBIT_ATTRBIT(&rattrbits, NFSATTRBIT_CHANGE); } } else error = NFSERR_SERVERFAULT; NFSUNLOCKCLSTATE(); } if (vp != NULL) vrele(vp); if (mp != NULL) vfs_unbusy(mp); if (nfhp != NULL) FREE((caddr_t)nfhp, M_NFSFH); if (!error) (void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va, NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0, (uint64_t)0); break; case NFSV4OP_CBRECALL: NFSCL_DEBUG(4, "cbrecall\n"); NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID + NFSX_UNSIGNED); stateid.seqid = *tl++; NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other, NFSX_STATEIDOTHER); tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); trunc = fxdr_unsigned(int, *tl); error = nfsm_getfh(nd, &nfhp); if (error == 0 && i == 0 && minorvers != NFSV4_MINORVERSION) error = NFSERR_OPNOTINSESS; if (!error) { NFSLOCKCLSTATE(); if (minorvers == NFSV4_MINORVERSION) clp = nfscl_getclnt(cbident); else clp = nfscl_getclntsess(sessionid); if (clp != NULL) { dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0) { dp->nfsdl_flags |= NFSCLDL_RECALL; wakeup((caddr_t)clp); } } else { error = NFSERR_SERVERFAULT; } NFSUNLOCKCLSTATE(); } if (nfhp != NULL) FREE((caddr_t)nfhp, M_NFSFH); break; case NFSV4OP_CBLAYOUTRECALL: NFSCL_DEBUG(4, "cblayrec\n"); nfhp = NULL; NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED); laytype = fxdr_unsigned(int, *tl++); iomode = fxdr_unsigned(uint32_t, *tl++); if (newnfs_true == *tl++) changed = 1; else changed = 0; recalltype = fxdr_unsigned(int, *tl); recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK); if (laytype != NFSLAYOUT_NFSV4_1_FILES) error = NFSERR_NOMATCHLAYOUT; else if (recalltype == NFSLAYOUTRETURN_FILE) { error = nfsm_getfh(nd, &nfhp); NFSCL_DEBUG(4, "retfile getfh=%d\n", error); if (error != 0) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID); off = fxdr_hyper(tl); tl += 2; len = fxdr_hyper(tl); tl += 2; stateid.seqid = fxdr_unsigned(uint32_t, *tl++); NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER); if (minorvers == NFSV4_MINORVERSION) error = NFSERR_NOTSUPP; else if (i == 0) error = NFSERR_OPNOTINSESS; if (error == 0) { NFSLOCKCLSTATE(); clp = nfscl_getclntsess(sessionid); NFSCL_DEBUG(4, "cbly clp=%p\n", clp); if (clp != NULL) { lyp = nfscl_findlayout(clp, nfhp->nfh_fh, nfhp->nfh_len); NFSCL_DEBUG(4, "cblyp=%p\n", lyp); if (lyp != NULL && (lyp->nfsly_flags & NFSLY_FILES) != 0 && !NFSBCMP(stateid.other, lyp->nfsly_stateid.other, NFSX_STATEIDOTHER)) { error = nfscl_layoutrecall( recalltype, lyp, iomode, off, len, stateid.seqid, recallp); recallp = NULL; wakeup(clp); NFSCL_DEBUG(4, "aft layrcal=%d\n", error); } else error = NFSERR_NOMATCHLAYOUT; } else error = NFSERR_NOMATCHLAYOUT; NFSUNLOCKCLSTATE(); } free(nfhp, M_NFSFH); } else if (recalltype == NFSLAYOUTRETURN_FSID) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER); filesid[0] = fxdr_hyper(tl); tl += 2; filesid[1] = fxdr_hyper(tl); tl += 2; gotone = 0; NFSLOCKCLSTATE(); clp = nfscl_getclntsess(sessionid); if (clp != NULL) { TAILQ_FOREACH(lyp, &clp->nfsc_layout, nfsly_list) { if (lyp->nfsly_filesid[0] == filesid[0] && lyp->nfsly_filesid[1] == filesid[1]) { error = nfscl_layoutrecall( recalltype, lyp, iomode, 0, UINT64_MAX, lyp->nfsly_stateid.seqid, recallp); recallp = NULL; gotone = 1; } } if (gotone != 0) wakeup(clp); else error = NFSERR_NOMATCHLAYOUT; } else error = NFSERR_NOMATCHLAYOUT; NFSUNLOCKCLSTATE(); } else if (recalltype == NFSLAYOUTRETURN_ALL) { gotone = 0; NFSLOCKCLSTATE(); clp = nfscl_getclntsess(sessionid); if (clp != NULL) { TAILQ_FOREACH(lyp, &clp->nfsc_layout, nfsly_list) { error = nfscl_layoutrecall( recalltype, lyp, iomode, 0, UINT64_MAX, lyp->nfsly_stateid.seqid, recallp); recallp = NULL; gotone = 1; } if (gotone != 0) wakeup(clp); else error = NFSERR_NOMATCHLAYOUT; } else error = NFSERR_NOMATCHLAYOUT; NFSUNLOCKCLSTATE(); } else error = NFSERR_NOMATCHLAYOUT; if (recallp != NULL) { free(recallp, M_NFSLAYRECALL); recallp = NULL; } break; case NFSV4OP_CBSEQUENCE: NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED); bcopy(tl, sessionid, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; seqid = fxdr_unsigned(uint32_t, *tl++); slotid = fxdr_unsigned(uint32_t, *tl++); highslot = fxdr_unsigned(uint32_t, *tl++); cachethis = *tl++; /* Throw away the referring call stuff. */ clist = fxdr_unsigned(int, *tl); for (j = 0; j < clist; j++) { NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID + NFSX_UNSIGNED); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; rcalls = fxdr_unsigned(int, *tl); for (k = 0; k < rcalls; k++) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED); } } NFSLOCKCLSTATE(); if (i == 0) { clp = nfscl_getclntsess(sessionid); if (clp == NULL) error = NFSERR_SERVERFAULT; } else error = NFSERR_SEQUENCEPOS; if (error == 0) { tsep = nfsmnt_mdssession(clp->nfsc_nmp); error = nfsv4_seqsession(seqid, slotid, highslot, tsep->nfsess_cbslots, &rep, tsep->nfsess_backslots); } NFSUNLOCKCLSTATE(); if (error == 0 || error == NFSERR_REPLYFROMCACHE) { gotseq_ok = 1; if (rep != NULL) { /* * Handle a reply for a retried * callback. The reply will be * re-inserted in the session cache * by the nfsv4_seqsess_cacherep() call * after out: */ KASSERT(error == NFSERR_REPLYFROMCACHE, ("cbsequence: non-NULL rep")); NFSCL_DEBUG(4, "Got cbretry\n"); m_freem(nd->nd_mreq); nd->nd_mreq = rep; rep = NULL; goto out; } NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED); bcopy(sessionid, tl, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; *tl++ = txdr_unsigned(seqid); *tl++ = txdr_unsigned(slotid); *tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1); *tl = txdr_unsigned(NFSV4_CBSLOTS - 1); } break; default: if (i == 0 && minorvers == NFSV41_MINORVERSION) error = NFSERR_OPNOTINSESS; else { NFSCL_DEBUG(1, "unsupp callback %d\n", op); error = NFSERR_NOTSUPP; } break; } if (error) { if (error == EBADRPC || error == NFSERR_BADXDR) { nd->nd_repstat = NFSERR_BADXDR; } else { nd->nd_repstat = error; } error = 0; } retops++; if (nd->nd_repstat) { *repp = nfscl_errmap(nd, minorvers); break; } else *repp = 0; /* NFS4_OK */ } nfsmout: if (recallp != NULL) free(recallp, M_NFSLAYRECALL); if (error) { if (error == EBADRPC || error == NFSERR_BADXDR) nd->nd_repstat = NFSERR_BADXDR; else printf("nfsv4 comperr1=%d\n", error); } if (taglen == -1) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = 0; *tl = 0; } else { *retopsp = txdr_unsigned(retops); } *nd->nd_errp = nfscl_errmap(nd, minorvers); out: if (gotseq_ok != 0) { rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK); NFSLOCKCLSTATE(); clp = nfscl_getclntsess(sessionid); if (clp != NULL) { tsep = nfsmnt_mdssession(clp->nfsc_nmp); nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots, NFSERR_OK, &rep); NFSUNLOCKCLSTATE(); } else { NFSUNLOCKCLSTATE(); m_freem(rep); } } } /* * Generate the next cbident value. Basically just increment a static value * and then check that it isn't already in the list, if it has wrapped around. */ static u_int32_t nfscl_nextcbident(void) { struct nfsclclient *clp; int matched; static u_int32_t nextcbident = 0; static int haswrapped = 0; nextcbident++; if (nextcbident == 0) haswrapped = 1; if (haswrapped) { /* * Search the clientid list for one already using this cbident. */ do { matched = 0; NFSLOCKCLSTATE(); LIST_FOREACH(clp, &nfsclhead, nfsc_list) { if (clp->nfsc_cbident == nextcbident) { matched = 1; break; } } NFSUNLOCKCLSTATE(); if (matched == 1) nextcbident++; } while (matched); } return (nextcbident); } /* * Get the mount point related to a given cbident or session and busy it. */ static mount_t nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident, struct nfsclclient **clpp) { struct nfsclclient *clp; mount_t mp; int error; struct nfsclsession *tsep; *clpp = NULL; NFSLOCKCLSTATE(); LIST_FOREACH(clp, &nfsclhead, nfsc_list) { tsep = nfsmnt_mdssession(clp->nfsc_nmp); if (minorvers == NFSV4_MINORVERSION) { if (clp->nfsc_cbident == cbident) break; } else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid, NFSX_V4SESSIONID)) break; } if (clp == NULL) { NFSUNLOCKCLSTATE(); return (NULL); } mp = clp->nfsc_nmp->nm_mountp; vfs_ref(mp); NFSUNLOCKCLSTATE(); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (NULL); *clpp = clp; return (mp); } /* * Get the clientid pointer related to a given cbident. */ static struct nfsclclient * nfscl_getclnt(u_int32_t cbident) { struct nfsclclient *clp; LIST_FOREACH(clp, &nfsclhead, nfsc_list) if (clp->nfsc_cbident == cbident) break; return (clp); } /* * Get the clientid pointer related to a given sessionid. */ static struct nfsclclient * nfscl_getclntsess(uint8_t *sessionid) { struct nfsclclient *clp; struct nfsclsession *tsep; LIST_FOREACH(clp, &nfsclhead, nfsc_list) { tsep = nfsmnt_mdssession(clp->nfsc_nmp); if (!NFSBCMP(tsep->nfsess_sessionid, sessionid, NFSX_V4SESSIONID)) break; } return (clp); } /* * Search for a lock conflict locally on the client. A conflict occurs if * - not same owner and overlapping byte range and at least one of them is * a write lock or this is an unlock. */ static int nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen, struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp, struct nfscllock **lopp) { struct nfsclowner *owp; struct nfsclopen *op; int ret; if (dp != NULL) { ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp); if (ret) return (ret); } LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if (op->nfso_fhlen == fhlen && !NFSBCMP(op->nfso_fh, fhp, fhlen)) { ret = nfscl_checkconflict(&op->nfso_lock, nlop, own, lopp); if (ret) return (ret); } } } return (0); } static int nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop, u_int8_t *own, struct nfscllock **lopp) { struct nfscllockowner *lp; struct nfscllock *lop; LIST_FOREACH(lp, lhp, nfsl_list) { if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) { LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) { if (lop->nfslo_first >= nlop->nfslo_end) break; if (lop->nfslo_end <= nlop->nfslo_first) continue; if (lop->nfslo_type == F_WRLCK || nlop->nfslo_type == F_WRLCK || nlop->nfslo_type == F_UNLCK) { if (lopp != NULL) *lopp = lop; return (NFSERR_DENIED); } } } } return (0); } /* * Check for a local conflicting lock. */ APPLESTATIC int nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off, u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags) { struct nfscllock *lop, nlck; struct nfscldeleg *dp; struct nfsnode *np; u_int8_t own[NFSV4CL_LOCKNAMELEN]; int error; nlck.nfslo_type = fl->l_type; nlck.nfslo_first = off; if (len == NFS64BITSSET) { nlck.nfslo_end = NFS64BITSSET; } else { nlck.nfslo_end = off + len; if (nlck.nfslo_end <= nlck.nfslo_first) return (NFSERR_INVAL); } np = VTONFS(vp); nfscl_filllockowner(id, own, flags); NFSLOCKCLSTATE(); dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, &nlck, own, dp, &lop); if (error != 0) { fl->l_whence = SEEK_SET; fl->l_start = lop->nfslo_first; if (lop->nfslo_end == NFS64BITSSET) fl->l_len = 0; else fl->l_len = lop->nfslo_end - lop->nfslo_first; fl->l_pid = (pid_t)0; fl->l_type = lop->nfslo_type; error = -1; /* no RPC required */ } else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) || fl->l_type == F_RDLCK)) { /* * The delegation ensures that there isn't a conflicting * lock on the server, so return -1 to indicate an RPC * isn't required. */ fl->l_type = F_UNLCK; error = -1; } NFSUNLOCKCLSTATE(); return (error); } /* * Handle Recall of a delegation. * The clp must be exclusive locked when this is called. */ static int nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp, struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p, int called_from_renewthread) { struct nfsclowner *owp, *lowp, *nowp; struct nfsclopen *op, *lop; struct nfscllockowner *lp; struct nfscllock *lckp; struct nfsnode *np; int error = 0, ret, gotvp = 0; if (vp == NULL) { /* * First, get a vnode for the file. This is needed to do RPCs. */ ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh, dp->nfsdl_fhlen, p, &np); if (ret) { /* * File isn't open, so nothing to move over to the * server. */ return (0); } vp = NFSTOV(np); gotvp = 1; } else { np = VTONFS(vp); } dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET; /* * Ok, if it's a write delegation, flush data to the server, so * that close/open consistency is retained. */ ret = 0; NFSLOCKNODE(np); if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) { np->n_flag |= NDELEGRECALL; NFSUNLOCKNODE(np); ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread); NFSLOCKNODE(np); np->n_flag &= ~NDELEGRECALL; } NFSINVALATTRCACHE(np); NFSUNLOCKNODE(np); if (ret == EIO && called_from_renewthread != 0) { /* * If the flush failed with EIO for the renew thread, * return now, so that the dirty buffer will be flushed * later. */ if (gotvp != 0) vrele(vp); return (ret); } /* * Now, for each openowner with opens issued locally, move them * over to state against the server. */ LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) { lop = LIST_FIRST(&lowp->nfsow_open); if (lop != NULL) { if (LIST_NEXT(lop, nfso_list) != NULL) panic("nfsdlg mult opens"); /* * Look for the same openowner against the server. */ LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) { if (!NFSBCMP(lowp->nfsow_owner, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) { newnfs_copycred(&dp->nfsdl_cred, cred); ret = nfscl_moveopen(vp, clp, nmp, lop, owp, dp, cred, p); if (ret == NFSERR_STALECLIENTID || ret == NFSERR_STALEDONTRECOVER || ret == NFSERR_BADSESSION) { if (gotvp) vrele(vp); return (ret); } if (ret) { nfscl_freeopen(lop, 1); if (!error) error = ret; } break; } } /* * If no openowner found, create one and get an open * for it. */ if (owp == NULL) { MALLOC(nowp, struct nfsclowner *, sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK); nfscl_newopen(clp, NULL, &owp, &nowp, &op, NULL, lowp->nfsow_owner, dp->nfsdl_fh, dp->nfsdl_fhlen, NULL, NULL); newnfs_copycred(&dp->nfsdl_cred, cred); ret = nfscl_moveopen(vp, clp, nmp, lop, owp, dp, cred, p); if (ret) { nfscl_freeopenowner(owp, 0); if (ret == NFSERR_STALECLIENTID || ret == NFSERR_STALEDONTRECOVER || ret == NFSERR_BADSESSION) { if (gotvp) vrele(vp); return (ret); } if (ret) { nfscl_freeopen(lop, 1); if (!error) error = ret; } } } } } /* * Now, get byte range locks for any locks done locally. */ LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) { newnfs_copycred(&dp->nfsdl_cred, cred); ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p); if (ret == NFSERR_STALESTATEID || ret == NFSERR_STALEDONTRECOVER || ret == NFSERR_STALECLIENTID || ret == NFSERR_BADSESSION) { if (gotvp) vrele(vp); return (ret); } if (ret && !error) error = ret; } } if (gotvp) vrele(vp); return (error); } /* * Move a locally issued open over to an owner on the state list. * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and * returns with it unlocked. */ static int nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp, struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp, struct ucred *cred, NFSPROC_T *p) { struct nfsclopen *op, *nop; struct nfscldeleg *ndp; struct nfsnode *np; int error = 0, newone; /* * First, look for an appropriate open, If found, just increment the * opencnt in it. */ LIST_FOREACH(op, &owp->nfsow_open, nfso_list) { if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode && op->nfso_fhlen == lop->nfso_fhlen && !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) { op->nfso_opencnt += lop->nfso_opencnt; nfscl_freeopen(lop, 1); return (0); } } /* No appropriate open, so we have to do one against the server. */ np = VTONFS(vp); MALLOC(nop, struct nfsclopen *, sizeof (struct nfsclopen) + lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK); newone = 0; nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner, lop->nfso_fh, lop->nfso_fhlen, cred, &newone); ndp = dp; error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen, lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op, NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p); if (error) { if (newone) nfscl_freeopen(op, 0); } else { op->nfso_mode |= lop->nfso_mode; op->nfso_opencnt += lop->nfso_opencnt; nfscl_freeopen(lop, 1); } if (nop != NULL) FREE((caddr_t)nop, M_NFSCLOPEN); if (ndp != NULL) { /* * What should I do with the returned delegation, since the * delegation is being recalled? For now, just printf and * through it away. */ printf("Moveopen returned deleg\n"); FREE((caddr_t)ndp, M_NFSCLDELEG); } return (error); } /* * Recall all delegations on this client. */ static void nfscl_totalrecall(struct nfsclclient *clp) { struct nfscldeleg *dp; TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) { if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0) dp->nfsdl_flags |= NFSCLDL_RECALL; } } /* * Relock byte ranges. Called for delegation recall and state expiry. */ static int nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp, struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred, NFSPROC_T *p) { struct nfscllockowner *nlp; struct nfsfh *nfhp; u_int64_t off, len; u_int32_t clidrev = 0; int error, newone, donelocally; off = lop->nfslo_first; len = lop->nfslo_end - lop->nfslo_first; error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p, clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner, lp->nfsl_openowner, &nlp, &newone, &donelocally); if (error || donelocally) return (error); if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; else clidrev = 0; nfhp = VTONFS(vp)->n_fhp; error = nfscl_trylock(nmp, vp, nfhp->nfh_fh, nfhp->nfh_len, nlp, newone, 0, off, len, lop->nfslo_type, cred, p); if (error) nfscl_freelockowner(nlp, 0); return (error); } /* * Called to re-open a file. Basically get a vnode for the file handle * and then call nfsrpc_openrpc() to do the rest. */ static int nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen, u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp, struct ucred *cred, NFSPROC_T *p) { struct nfsnode *np; vnode_t vp; int error; error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np); if (error) return (error); vp = NFSTOV(np); if (np->n_v4 != NULL) { error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen, fhp, fhlen, mode, op, NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0, cred, p); } else { error = EINVAL; } vrele(vp); return (error); } /* * Try an open against the server. Just call nfsrpc_openrpc(), retrying while * NFSERR_DELAY. Also, try system credentials, if the passed in credentials * fail. */ static int nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen, u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op, u_int8_t *name, int namelen, struct nfscldeleg **ndpp, int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p) { int error; do { error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen, mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p, 0, 0); if (error == NFSERR_DELAY) (void) nfs_catnap(PZERO, error, "nfstryop"); } while (error == NFSERR_DELAY); if (error == EAUTH || error == EACCES) { /* Try again using system credentials */ newnfs_setroot(cred); do { error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen, mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p, 1, 0); if (error == NFSERR_DELAY) (void) nfs_catnap(PZERO, error, "nfstryop"); } while (error == NFSERR_DELAY); } return (error); } /* * Try a byte range lock. Just loop on nfsrpc_lock() while it returns * NFSERR_DELAY. Also, retry with system credentials, if the provided * cred don't work. */ static int nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen, struct nfscllockowner *nlp, int newone, int reclaim, u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p) { struct nfsrv_descript nfsd, *nd = &nfsd; int error; do { error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone, reclaim, off, len, type, cred, p, 0); if (!error && nd->nd_repstat == NFSERR_DELAY) (void) nfs_catnap(PZERO, (int)nd->nd_repstat, "nfstrylck"); } while (!error && nd->nd_repstat == NFSERR_DELAY); if (!error) error = nd->nd_repstat; if (error == EAUTH || error == EACCES) { /* Try again using root credentials */ newnfs_setroot(cred); do { error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone, reclaim, off, len, type, cred, p, 1); if (!error && nd->nd_repstat == NFSERR_DELAY) (void) nfs_catnap(PZERO, (int)nd->nd_repstat, "nfstrylck"); } while (!error && nd->nd_repstat == NFSERR_DELAY); if (!error) error = nd->nd_repstat; } return (error); } /* * Try a delegreturn against the server. Just call nfsrpc_delegreturn(), * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in * credentials fail. */ static int nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred, struct nfsmount *nmp, NFSPROC_T *p) { int error; do { error = nfsrpc_delegreturn(dp, cred, nmp, p, 0); if (error == NFSERR_DELAY) (void) nfs_catnap(PZERO, error, "nfstrydp"); } while (error == NFSERR_DELAY); if (error == EAUTH || error == EACCES) { /* Try again using system credentials */ newnfs_setroot(cred); do { error = nfsrpc_delegreturn(dp, cred, nmp, p, 1); if (error == NFSERR_DELAY) (void) nfs_catnap(PZERO, error, "nfstrydp"); } while (error == NFSERR_DELAY); } return (error); } /* * Try a close against the server. Just call nfsrpc_closerpc(), * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in * credentials fail. */ APPLESTATIC int nfscl_tryclose(struct nfsclopen *op, struct ucred *cred, struct nfsmount *nmp, NFSPROC_T *p) { struct nfsrv_descript nfsd, *nd = &nfsd; int error; do { error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0); if (error == NFSERR_DELAY) (void) nfs_catnap(PZERO, error, "nfstrycl"); } while (error == NFSERR_DELAY); if (error == EAUTH || error == EACCES) { /* Try again using system credentials */ newnfs_setroot(cred); do { error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1); if (error == NFSERR_DELAY) (void) nfs_catnap(PZERO, error, "nfstrycl"); } while (error == NFSERR_DELAY); } return (error); } /* * Decide if a delegation on a file permits close without flushing writes * to the server. This might be a big performance win in some environments. * (Not useful until the client does caching on local stable storage.) */ APPLESTATIC int nfscl_mustflush(vnode_t vp) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsnode *np; struct nfsmount *nmp; np = VTONFS(vp); nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return (1); NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return (1); } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == NFSCLDL_WRITE && (dp->nfsdl_sizelimit >= np->n_size || !NFSHASSTRICT3530(nmp))) { NFSUNLOCKCLSTATE(); return (0); } NFSUNLOCKCLSTATE(); return (1); } /* * See if a (write) delegation exists for this file. */ APPLESTATIC int nfscl_nodeleg(vnode_t vp, int writedeleg) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsnode *np; struct nfsmount *nmp; np = VTONFS(vp); nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return (1); NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return (1); } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 && (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) == NFSCLDL_WRITE)) { NFSUNLOCKCLSTATE(); return (0); } NFSUNLOCKCLSTATE(); return (1); } /* * Look for an associated delegation that should be DelegReturned. */ APPLESTATIC int nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsclowner *owp; struct nfscllockowner *lp; struct nfsmount *nmp; struct ucred *cred; struct nfsnode *np; int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept; nmp = VFSTONFS(vnode_mount(vp)); np = VTONFS(vp); NFSLOCKCLSTATE(); /* * Loop around waiting for: * - outstanding I/O operations on delegations to complete * - for a delegation on vp that has state, lock the client and * do a recall * - return delegation with no state */ while (1) { clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return (retcnt); } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL) { /* * Wait for outstanding I/O ops to be done. */ if (dp->nfsdl_rwlock.nfslock_usecnt > 0) { if (igotlock) { nfsv4_unlock(&clp->nfsc_lock, 0); igotlock = 0; } dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED; (void) nfsmsleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL); continue; } needsrecall = 0; LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) { if (!LIST_EMPTY(&owp->nfsow_open)) { needsrecall = 1; break; } } if (!needsrecall) { LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { if (!LIST_EMPTY(&lp->nfsl_lock)) { needsrecall = 1; break; } } } if (needsrecall && !triedrecall) { dp->nfsdl_flags |= NFSCLDL_DELEGRET; islept = 0; while (!igotlock) { igotlock = nfsv4_lock(&clp->nfsc_lock, 1, &islept, NFSCLSTATEMUTEXPTR, NULL); if (islept) break; } if (islept) continue; NFSUNLOCKCLSTATE(); cred = newnfs_getcred(); newnfs_copycred(&dp->nfsdl_cred, cred); (void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0); NFSFREECRED(cred); triedrecall = 1; NFSLOCKCLSTATE(); nfsv4_unlock(&clp->nfsc_lock, 0); igotlock = 0; continue; } *stp = dp->nfsdl_stateid; retcnt = 1; nfscl_cleandeleg(dp); nfscl_freedeleg(&clp->nfsc_deleg, dp); } if (igotlock) nfsv4_unlock(&clp->nfsc_lock, 0); NFSUNLOCKCLSTATE(); return (retcnt); } } /* * Look for associated delegation(s) that should be DelegReturned. */ APPLESTATIC int nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp, nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsclowner *owp; struct nfscllockowner *lp; struct nfsmount *nmp; struct ucred *cred; struct nfsnode *np; int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept; nmp = VFSTONFS(vnode_mount(fvp)); *gotfdp = 0; *gottdp = 0; NFSLOCKCLSTATE(); /* * Loop around waiting for: * - outstanding I/O operations on delegations to complete * - for a delegation on fvp that has state, lock the client and * do a recall * - return delegation(s) with no state. */ while (1) { clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return (retcnt); } np = VTONFS(fvp); dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && *gotfdp == 0) { /* * Wait for outstanding I/O ops to be done. */ if (dp->nfsdl_rwlock.nfslock_usecnt > 0) { if (igotlock) { nfsv4_unlock(&clp->nfsc_lock, 0); igotlock = 0; } dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED; (void) nfsmsleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL); continue; } needsrecall = 0; LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) { if (!LIST_EMPTY(&owp->nfsow_open)) { needsrecall = 1; break; } } if (!needsrecall) { LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { if (!LIST_EMPTY(&lp->nfsl_lock)) { needsrecall = 1; break; } } } if (needsrecall && !triedrecall) { dp->nfsdl_flags |= NFSCLDL_DELEGRET; islept = 0; while (!igotlock) { igotlock = nfsv4_lock(&clp->nfsc_lock, 1, &islept, NFSCLSTATEMUTEXPTR, NULL); if (islept) break; } if (islept) continue; NFSUNLOCKCLSTATE(); cred = newnfs_getcred(); newnfs_copycred(&dp->nfsdl_cred, cred); (void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0); NFSFREECRED(cred); triedrecall = 1; NFSLOCKCLSTATE(); nfsv4_unlock(&clp->nfsc_lock, 0); igotlock = 0; continue; } *fstp = dp->nfsdl_stateid; retcnt++; *gotfdp = 1; nfscl_cleandeleg(dp); nfscl_freedeleg(&clp->nfsc_deleg, dp); } if (igotlock) { nfsv4_unlock(&clp->nfsc_lock, 0); igotlock = 0; } if (tvp != NULL) { np = VTONFS(tvp); dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && *gottdp == 0) { /* * Wait for outstanding I/O ops to be done. */ if (dp->nfsdl_rwlock.nfslock_usecnt > 0) { dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED; (void) nfsmsleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL); continue; } LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) { if (!LIST_EMPTY(&owp->nfsow_open)) { NFSUNLOCKCLSTATE(); return (retcnt); } } LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) { if (!LIST_EMPTY(&lp->nfsl_lock)) { NFSUNLOCKCLSTATE(); return (retcnt); } } *tstp = dp->nfsdl_stateid; retcnt++; *gottdp = 1; nfscl_cleandeleg(dp); nfscl_freedeleg(&clp->nfsc_deleg, dp); } } NFSUNLOCKCLSTATE(); return (retcnt); } } /* * Get a reference on the clientid associated with the mount point. * Return 1 if success, 0 otherwise. */ APPLESTATIC int nfscl_getref(struct nfsmount *nmp) { struct nfsclclient *clp; NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return (0); } nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL); NFSUNLOCKCLSTATE(); return (1); } /* * Release a reference on a clientid acquired with the above call. */ APPLESTATIC void nfscl_relref(struct nfsmount *nmp) { struct nfsclclient *clp; NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return; } nfsv4_relref(&clp->nfsc_lock); NFSUNLOCKCLSTATE(); } /* * Save the size attribute in the delegation, since the nfsnode * is going away. */ APPLESTATIC void nfscl_reclaimnode(vnode_t vp) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp; nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return; NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return; } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) dp->nfsdl_size = np->n_size; NFSUNLOCKCLSTATE(); } /* * Get the saved size attribute in the delegation, since it is a * newly allocated nfsnode. */ APPLESTATIC void nfscl_newnode(vnode_t vp) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp; nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return; NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return; } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) np->n_size = dp->nfsdl_size; NFSUNLOCKCLSTATE(); } /* * If there is a valid write delegation for this file, set the modtime * to the local clock time. */ APPLESTATIC void nfscl_delegmodtime(vnode_t vp) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp; nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return; NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return; } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) { nanotime(&dp->nfsdl_modtime); dp->nfsdl_flags |= NFSCLDL_MODTIMESET; } NFSUNLOCKCLSTATE(); } /* * If there is a valid write delegation for this file with a modtime set, * put that modtime in mtime. */ APPLESTATIC void nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime) { struct nfsclclient *clp; struct nfscldeleg *dp; struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp; nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return; NFSLOCKCLSTATE(); clp = nfscl_findcl(nmp); if (clp == NULL) { NFSUNLOCKCLSTATE(); return; } dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (dp != NULL && (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) == (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) *mtime = dp->nfsdl_modtime; NFSUNLOCKCLSTATE(); } static int nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers) { short *defaulterrp, *errp; if (!nd->nd_repstat) return (0); if (nd->nd_procnum == NFSPROC_NOOP) return (txdr_unsigned(nd->nd_repstat & 0xffff)); if (nd->nd_repstat == EBADRPC) return (txdr_unsigned(NFSERR_BADXDR)); if (nd->nd_repstat == NFSERR_MINORVERMISMATCH || nd->nd_repstat == NFSERR_OPILLEGAL) return (txdr_unsigned(nd->nd_repstat)); if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 && minorvers > NFSV4_MINORVERSION) { /* NFSv4.n error. */ return (txdr_unsigned(nd->nd_repstat)); } if (nd->nd_procnum < NFSV4OP_CBNOPS) errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum]; else return (txdr_unsigned(nd->nd_repstat)); while (*++errp) if (*errp == (short)nd->nd_repstat) return (txdr_unsigned(nd->nd_repstat)); return (txdr_unsigned(*defaulterrp)); } /* * Called to find/add a layout to a client. * This function returns the layout with a refcnt (shared lock) upon * success (returns 0) or with no lock/refcnt on the layout when an * error is returned. * If a layout is passed in via lypp, it is locked (exclusively locked). */ APPLESTATIC int nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen, nfsv4stateid_t *stateidp, int retonclose, struct nfsclflayouthead *fhlp, struct nfscllayout **lypp, struct ucred *cred, NFSPROC_T *p) { struct nfsclclient *clp; struct nfscllayout *lyp, *tlyp; struct nfsclflayout *flp; struct nfsnode *np = VTONFS(vp); mount_t mp; int layout_passed_in; mp = nmp->nm_mountp; layout_passed_in = 1; tlyp = NULL; lyp = *lypp; if (lyp == NULL) { layout_passed_in = 0; tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT, M_WAITOK | M_ZERO); } NFSLOCKCLSTATE(); clp = nmp->nm_clp; if (clp == NULL) { if (layout_passed_in != 0) nfsv4_unlock(&lyp->nfsly_lock, 0); NFSUNLOCKCLSTATE(); if (tlyp != NULL) free(tlyp, M_NFSLAYOUT); return (EPERM); } if (lyp == NULL) { /* * Although no lyp was passed in, another thread might have * allocated one. If one is found, just increment it's ref * count and return it. */ lyp = nfscl_findlayout(clp, fhp, fhlen); if (lyp == NULL) { lyp = tlyp; tlyp = NULL; lyp->nfsly_stateid.seqid = stateidp->seqid; lyp->nfsly_stateid.other[0] = stateidp->other[0]; lyp->nfsly_stateid.other[1] = stateidp->other[1]; lyp->nfsly_stateid.other[2] = stateidp->other[2]; lyp->nfsly_lastbyte = 0; LIST_INIT(&lyp->nfsly_flayread); LIST_INIT(&lyp->nfsly_flayrw); LIST_INIT(&lyp->nfsly_recall); lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0]; lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1]; lyp->nfsly_clp = clp; lyp->nfsly_flags = (retonclose != 0) ? (NFSLY_FILES | NFSLY_RETONCLOSE) : NFSLY_FILES; lyp->nfsly_fhlen = fhlen; NFSBCOPY(fhp, lyp->nfsly_fh, fhlen); TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list); LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp, nfsly_hash); lyp->nfsly_timestamp = NFSD_MONOSEC + 120; nfscl_layoutcnt++; } else { if (retonclose != 0) lyp->nfsly_flags |= NFSLY_RETONCLOSE; TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list); TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list); lyp->nfsly_timestamp = NFSD_MONOSEC + 120; } nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp); - if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (NFSCL_FORCEDISM(mp)) { NFSUNLOCKCLSTATE(); if (tlyp != NULL) free(tlyp, M_NFSLAYOUT); return (EPERM); } *lypp = lyp; } else lyp->nfsly_stateid.seqid = stateidp->seqid; /* Merge the new list of File Layouts into the list. */ flp = LIST_FIRST(fhlp); if (flp != NULL) { if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ) nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp); else nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp); } if (layout_passed_in != 0) nfsv4_unlock(&lyp->nfsly_lock, 1); NFSUNLOCKCLSTATE(); if (tlyp != NULL) free(tlyp, M_NFSLAYOUT); return (0); } /* * Search for a layout by MDS file handle. * If one is found, it is returned with a refcnt (shared lock) iff * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is * returned NULL. */ struct nfscllayout * nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen, uint64_t off, struct nfsclflayout **retflpp, int *recalledp) { struct nfscllayout *lyp; mount_t mp; int error, igotlock; mp = clp->nfsc_nmp->nm_mountp; *recalledp = 0; *retflpp = NULL; NFSLOCKCLSTATE(); lyp = nfscl_findlayout(clp, fhp, fhlen); if (lyp != NULL) { if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) { TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list); TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list); lyp->nfsly_timestamp = NFSD_MONOSEC + 120; error = nfscl_findlayoutforio(lyp, off, NFSV4OPEN_ACCESSREAD, retflpp); if (error == 0) nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp); else { do { igotlock = nfsv4_lock(&lyp->nfsly_lock, 1, NULL, NFSCLSTATEMUTEXPTR, mp); - } while (igotlock == 0 && - (mp->mnt_kern_flag & MNTK_UNMOUNTF) == 0); + } while (igotlock == 0 && !NFSCL_FORCEDISM(mp)); *retflpp = NULL; } - if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (NFSCL_FORCEDISM(mp)) { lyp = NULL; *recalledp = 1; } } else { lyp = NULL; *recalledp = 1; } } NFSUNLOCKCLSTATE(); return (lyp); } /* * Search for a layout by MDS file handle. If one is found, mark in to be * recalled, if it already marked "return on close". */ static void nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp, int fhlen, struct nfsclrecalllayout **recallpp) { struct nfscllayout *lyp; uint32_t iomode; if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) || nfscl_enablecallb == 0 || nfs_numnfscbd == 0 || (VTONFS(vp)->n_flag & NNOLAYOUT) != 0) return; lyp = nfscl_findlayout(clp, fhp, fhlen); if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE | NFSLY_RECALL)) == NFSLY_RETONCLOSE) { iomode = 0; if (!LIST_EMPTY(&lyp->nfsly_flayread)) iomode |= NFSLAYOUTIOMODE_READ; if (!LIST_EMPTY(&lyp->nfsly_flayrw)) iomode |= NFSLAYOUTIOMODE_RW; (void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode, 0, UINT64_MAX, lyp->nfsly_stateid.seqid, *recallpp); NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode); *recallpp = NULL; } } /* * Dereference a layout. */ void nfscl_rellayout(struct nfscllayout *lyp, int exclocked) { NFSLOCKCLSTATE(); if (exclocked != 0) nfsv4_unlock(&lyp->nfsly_lock, 0); else nfsv4_relref(&lyp->nfsly_lock); NFSUNLOCKCLSTATE(); } /* * Search for a devinfo by deviceid. If one is found, return it after * acquiring a reference count on it. */ struct nfscldevinfo * nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid, struct nfscldevinfo *dip) { NFSLOCKCLSTATE(); if (dip == NULL) dip = nfscl_finddevinfo(clp, deviceid); if (dip != NULL) dip->nfsdi_refcnt++; NFSUNLOCKCLSTATE(); return (dip); } /* * Dereference a devinfo structure. */ static void nfscl_reldevinfo_locked(struct nfscldevinfo *dip) { dip->nfsdi_refcnt--; if (dip->nfsdi_refcnt == 0) wakeup(&dip->nfsdi_refcnt); } /* * Dereference a devinfo structure. */ void nfscl_reldevinfo(struct nfscldevinfo *dip) { NFSLOCKCLSTATE(); nfscl_reldevinfo_locked(dip); NFSUNLOCKCLSTATE(); } /* * Find a layout for this file handle. Return NULL upon failure. */ static struct nfscllayout * nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen) { struct nfscllayout *lyp; LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash) if (lyp->nfsly_fhlen == fhlen && !NFSBCMP(lyp->nfsly_fh, fhp, fhlen)) break; return (lyp); } /* * Find a devinfo for this deviceid. Return NULL upon failure. */ static struct nfscldevinfo * nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid) { struct nfscldevinfo *dip; LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list) if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID) == 0) break; return (dip); } /* * Merge the new file layout list into the main one, maintaining it in * increasing offset order. */ static void nfscl_mergeflayouts(struct nfsclflayouthead *fhlp, struct nfsclflayouthead *newfhlp) { struct nfsclflayout *flp, *nflp, *prevflp, *tflp; flp = LIST_FIRST(fhlp); prevflp = NULL; LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) { while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) { prevflp = flp; flp = LIST_NEXT(flp, nfsfl_list); } if (prevflp == NULL) LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list); else LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list); prevflp = nflp; } } /* * Add this nfscldevinfo to the client, if it doesn't already exist. * This function consumes the structure pointed at by dip, if not NULL. */ APPLESTATIC int nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, struct nfsclflayout *flp) { struct nfsclclient *clp; struct nfscldevinfo *tdip; NFSLOCKCLSTATE(); clp = nmp->nm_clp; if (clp == NULL) { NFSUNLOCKCLSTATE(); if (dip != NULL) free(dip, M_NFSDEVINFO); return (ENODEV); } tdip = nfscl_finddevinfo(clp, flp->nfsfl_dev); if (tdip != NULL) { tdip->nfsdi_layoutrefs++; flp->nfsfl_devp = tdip; nfscl_reldevinfo_locked(tdip); NFSUNLOCKCLSTATE(); if (dip != NULL) free(dip, M_NFSDEVINFO); return (0); } if (dip != NULL) { LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list); dip->nfsdi_layoutrefs = 1; flp->nfsfl_devp = dip; } NFSUNLOCKCLSTATE(); if (dip == NULL) return (ENODEV); return (0); } /* * Free up a layout structure and associated file layout structure(s). */ APPLESTATIC void nfscl_freelayout(struct nfscllayout *layp) { struct nfsclflayout *flp, *nflp; struct nfsclrecalllayout *rp, *nrp; LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) { LIST_REMOVE(flp, nfsfl_list); nfscl_freeflayout(flp); } LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) { LIST_REMOVE(flp, nfsfl_list); nfscl_freeflayout(flp); } LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) { LIST_REMOVE(rp, nfsrecly_list); free(rp, M_NFSLAYRECALL); } nfscl_layoutcnt--; free(layp, M_NFSLAYOUT); } /* * Free up a file layout structure. */ APPLESTATIC void nfscl_freeflayout(struct nfsclflayout *flp) { int i; for (i = 0; i < flp->nfsfl_fhcnt; i++) free(flp->nfsfl_fh[i], M_NFSFH); if (flp->nfsfl_devp != NULL) flp->nfsfl_devp->nfsdi_layoutrefs--; free(flp, M_NFSFLAYOUT); } /* * Free up a file layout devinfo structure. */ APPLESTATIC void nfscl_freedevinfo(struct nfscldevinfo *dip) { free(dip, M_NFSDEVINFO); } /* * Mark any layouts that match as recalled. */ static int nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode, uint64_t off, uint64_t len, uint32_t stateseqid, struct nfsclrecalllayout *recallp) { struct nfsclrecalllayout *rp, *orp; recallp->nfsrecly_recalltype = recalltype; recallp->nfsrecly_iomode = iomode; recallp->nfsrecly_stateseqid = stateseqid; recallp->nfsrecly_off = off; recallp->nfsrecly_len = len; /* * Order the list as file returns first, followed by fsid and any * returns, both in increasing stateseqid order. * Note that the seqids wrap around, so 1 is after 0xffffffff. * (I'm not sure this is correct because I find RFC5661 confusing * on this, but hopefully it will work ok.) */ orp = NULL; LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) { orp = rp; if ((recalltype == NFSLAYOUTRETURN_FILE && (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE || nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) || (recalltype != NFSLAYOUTRETURN_FILE && rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE && nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) { LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list); break; } } if (rp == NULL) { if (orp == NULL) LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp, nfsrecly_list); else LIST_INSERT_AFTER(orp, recallp, nfsrecly_list); } lyp->nfsly_flags |= NFSLY_RECALL; return (0); } /* * Compare the two seqids for ordering. The trick is that the seqids can * wrap around from 0xffffffff->0, so check for the cases where one * has wrapped around. * Return 1 if seqid1 comes before seqid2, 0 otherwise. */ static int nfscl_seq(uint32_t seqid1, uint32_t seqid2) { if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff) /* seqid2 has wrapped around. */ return (0); if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff) /* seqid1 has wrapped around. */ return (1); if (seqid1 <= seqid2) return (1); return (0); } /* * Do a layout return for each of the recalls. */ static void nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp, struct ucred *cred, NFSPROC_T *p) { struct nfsclrecalllayout *rp; nfsv4stateid_t stateid; NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER); stateid.seqid = lyp->nfsly_stateid.seqid; LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) { (void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh, lyp->nfsly_fhlen, 0, NFSLAYOUT_NFSV4_1_FILES, rp->nfsrecly_iomode, rp->nfsrecly_recalltype, rp->nfsrecly_off, rp->nfsrecly_len, &stateid, 0, NULL, cred, p, NULL); } } /* * Do the layout commit for a file layout. */ static void nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp, struct ucred *cred, NFSPROC_T *p) { struct nfsclflayout *flp; uint64_t len; int error; LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) { if (flp->nfsfl_off <= lyp->nfsly_lastbyte) { len = flp->nfsfl_end - flp->nfsfl_off; error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh, lyp->nfsly_fhlen, 0, flp->nfsfl_off, len, lyp->nfsly_lastbyte, &lyp->nfsly_stateid, NFSLAYOUT_NFSV4_1_FILES, 0, NULL, cred, p, NULL); NFSCL_DEBUG(4, "layoutcommit err=%d\n", error); if (error == NFSERR_NOTSUPP) { /* If not supported, don't bother doing it. */ NFSLOCKMNT(nmp); nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT; NFSUNLOCKMNT(nmp); break; } } } } /* * Commit all layouts for a file (vnode). */ int nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p) { struct nfsclclient *clp; struct nfscllayout *lyp; struct nfsnode *np = VTONFS(vp); mount_t mp; struct nfsmount *nmp; mp = vnode_mount(vp); nmp = VFSTONFS(mp); if (NFSHASNOLAYOUTCOMMIT(nmp)) return (0); NFSLOCKCLSTATE(); clp = nmp->nm_clp; if (clp == NULL) { NFSUNLOCKCLSTATE(); return (EPERM); } lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len); if (lyp == NULL) { NFSUNLOCKCLSTATE(); return (EPERM); } nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp); - if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (NFSCL_FORCEDISM(mp)) { NFSUNLOCKCLSTATE(); return (EPERM); } tryagain: if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) { lyp->nfsly_flags &= ~NFSLY_WRITTEN; NFSUNLOCKCLSTATE(); NFSCL_DEBUG(4, "do layoutcommit2\n"); nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p); NFSLOCKCLSTATE(); goto tryagain; } nfsv4_relref(&lyp->nfsly_lock); NFSUNLOCKCLSTATE(); return (0); } Index: head/sys/fs/nfsclient/nfs_clvfsops.c =================================================================== --- head/sys/fs/nfsclient/nfs_clvfsops.c (revision 321627) +++ head/sys/fs/nfsclient/nfs_clvfsops.c (revision 321628) @@ -1,2025 +1,2025 @@ /*- * Copyright (c) 1989, 1993, 1995 * 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. * * from nfs_vfsops.c 8.12 (Berkeley) 5/20/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_bootp.h" #include "opt_nfsroot.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 #include #include #include #include #include #include FEATURE(nfscl, "NFSv4 client"); extern int nfscl_ticks; extern struct timeval nfsboottime; extern int nfsrv_useacl; extern int nfscl_debuglevel; extern enum nfsiod_state ncl_iodwant[NFS_MAXASYNCDAEMON]; extern struct nfsmount *ncl_iodmount[NFS_MAXASYNCDAEMON]; extern struct mtx ncl_iod_mutex; NFSCLSTATEMUTEX; MALLOC_DEFINE(M_NEWNFSREQ, "newnfsclient_req", "NFS request header"); MALLOC_DEFINE(M_NEWNFSMNT, "newnfsmnt", "NFS mount struct"); SYSCTL_DECL(_vfs_nfs); static int nfs_ip_paranoia = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, &nfs_ip_paranoia, 0, ""); static int nfs_tprintf_initial_delay = NFS_TPRINTF_INITIAL_DELAY; SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_INITIAL_DELAY, downdelayinitial, CTLFLAG_RW, &nfs_tprintf_initial_delay, 0, ""); /* how long between console messages "nfs server foo not responding" */ static int nfs_tprintf_delay = NFS_TPRINTF_DELAY; SYSCTL_INT(_vfs_nfs, NFS_TPRINTF_DELAY, downdelayinterval, CTLFLAG_RW, &nfs_tprintf_delay, 0, ""); #ifdef NFS_DEBUG int nfs_debug; SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "Toggle debug flag"); #endif static int nfs_mountroot(struct mount *); static void nfs_sec_name(char *, int *); static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp, const char *, struct ucred *, struct thread *); static int mountnfs(struct nfs_args *, struct mount *, struct sockaddr *, char *, u_char *, int, u_char *, int, u_char *, int, struct vnode **, struct ucred *, struct thread *, int, int, int); static void nfs_getnlminfo(struct vnode *, uint8_t *, size_t *, struct sockaddr_storage *, int *, off_t *, struct timeval *); static vfs_mount_t nfs_mount; static vfs_cmount_t nfs_cmount; static vfs_unmount_t nfs_unmount; static vfs_root_t nfs_root; static vfs_statfs_t nfs_statfs; static vfs_sync_t nfs_sync; static vfs_sysctl_t nfs_sysctl; static vfs_purge_t nfs_purge; /* * nfs vfs operations. */ static struct vfsops nfs_vfsops = { .vfs_init = ncl_init, .vfs_mount = nfs_mount, .vfs_cmount = nfs_cmount, .vfs_root = nfs_root, .vfs_statfs = nfs_statfs, .vfs_sync = nfs_sync, .vfs_uninit = ncl_uninit, .vfs_unmount = nfs_unmount, .vfs_sysctl = nfs_sysctl, .vfs_purge = nfs_purge, }; VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK | VFCF_SBDRY); /* So that loader and kldload(2) can find us, wherever we are.. */ MODULE_VERSION(nfs, 1); MODULE_DEPEND(nfs, nfscommon, 1, 1, 1); MODULE_DEPEND(nfs, krpc, 1, 1, 1); MODULE_DEPEND(nfs, nfssvc, 1, 1, 1); MODULE_DEPEND(nfs, nfslock, 1, 1, 1); /* * This structure is now defined in sys/nfs/nfs_diskless.c so that it * can be shared by both NFS clients. It is declared here so that it * will be defined for kernels built without NFS_ROOT, although it * isn't used in that case. */ #if !defined(NFS_ROOT) struct nfs_diskless nfs_diskless = { { { 0 } } }; struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; int nfs_diskless_valid = 0; #endif SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD, &nfs_diskless_valid, 0, "Has the diskless struct been filled correctly"); SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, nfsv3_diskless.root_hostnam, 0, "Path to nfs root"); SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, &nfsv3_diskless.root_saddr, sizeof(nfsv3_diskless.root_saddr), "%Ssockaddr_in", "Diskless root nfs address"); void newnfsargs_ntoh(struct nfs_args *); static int nfs_mountdiskless(char *, struct sockaddr_in *, struct nfs_args *, struct thread *, struct vnode **, struct mount *); static void nfs_convert_diskless(void); static void nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs); int newnfs_iosize(struct nfsmount *nmp) { int iosize, maxio; /* First, set the upper limit for iosize */ if (nmp->nm_flag & NFSMNT_NFSV4) { maxio = NFS_MAXBSIZE; } else if (nmp->nm_flag & NFSMNT_NFSV3) { if (nmp->nm_sotype == SOCK_DGRAM) maxio = NFS_MAXDGRAMDATA; else maxio = NFS_MAXBSIZE; } else { maxio = NFS_V2MAXDATA; } if (nmp->nm_rsize > maxio || nmp->nm_rsize == 0) nmp->nm_rsize = maxio; if (nmp->nm_rsize > NFS_MAXBSIZE) nmp->nm_rsize = NFS_MAXBSIZE; if (nmp->nm_readdirsize > maxio || nmp->nm_readdirsize == 0) nmp->nm_readdirsize = maxio; if (nmp->nm_readdirsize > nmp->nm_rsize) nmp->nm_readdirsize = nmp->nm_rsize; if (nmp->nm_wsize > maxio || nmp->nm_wsize == 0) nmp->nm_wsize = maxio; if (nmp->nm_wsize > NFS_MAXBSIZE) nmp->nm_wsize = NFS_MAXBSIZE; /* * Calculate the size used for io buffers. Use the larger * of the two sizes to minimise nfs requests but make sure * that it is at least one VM page to avoid wasting buffer * space. It must also be at least NFS_DIRBLKSIZ, since * that is the buffer size used for directories. */ iosize = imax(nmp->nm_rsize, nmp->nm_wsize); iosize = imax(iosize, PAGE_SIZE); iosize = imax(iosize, NFS_DIRBLKSIZ); nmp->nm_mountp->mnt_stat.f_iosize = iosize; return (iosize); } static void nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs) { args->version = NFS_ARGSVERSION; args->addr = oargs->addr; args->addrlen = oargs->addrlen; args->sotype = oargs->sotype; args->proto = oargs->proto; args->fh = oargs->fh; args->fhsize = oargs->fhsize; args->flags = oargs->flags; args->wsize = oargs->wsize; args->rsize = oargs->rsize; args->readdirsize = oargs->readdirsize; args->timeo = oargs->timeo; args->retrans = oargs->retrans; args->readahead = oargs->readahead; args->hostname = oargs->hostname; } static void nfs_convert_diskless(void) { bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, sizeof(struct ifaliasreq)); bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, sizeof(struct sockaddr_in)); nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); if (nfsv3_diskless.root_args.flags & NFSMNT_NFSV3) { nfsv3_diskless.root_fhsize = NFSX_MYFH; bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_MYFH); } else { nfsv3_diskless.root_fhsize = NFSX_V2FH; bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH); } bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, sizeof(struct sockaddr_in)); bcopy(nfs_diskless.root_hostnam, nfsv3_diskless.root_hostnam, MNAMELEN); nfsv3_diskless.root_time = nfs_diskless.root_time; bcopy(nfs_diskless.my_hostnam, nfsv3_diskless.my_hostnam, MAXHOSTNAMELEN); nfs_diskless_valid = 3; } /* * nfs statfs call */ static int nfs_statfs(struct mount *mp, struct statfs *sbp) { struct vnode *vp; struct thread *td; struct nfsmount *nmp = VFSTONFS(mp); struct nfsvattr nfsva; struct nfsfsinfo fs; struct nfsstatfs sb; int error = 0, attrflag, gotfsinfo = 0, ret; struct nfsnode *np; td = curthread; error = vfs_busy(mp, MBF_NOWAIT); if (error) return (error); error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); if (error) { vfs_unbusy(mp); return (error); } vp = NFSTOV(np); mtx_lock(&nmp->nm_mtx); if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) { mtx_unlock(&nmp->nm_mtx); error = nfsrpc_fsinfo(vp, &fs, td->td_ucred, td, &nfsva, &attrflag, NULL); if (!error) gotfsinfo = 1; } else mtx_unlock(&nmp->nm_mtx); if (!error) error = nfsrpc_statfs(vp, &sb, &fs, td->td_ucred, td, &nfsva, &attrflag, NULL); if (error != 0) NFSCL_DEBUG(2, "statfs=%d\n", error); if (attrflag == 0) { ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1, td->td_ucred, td, &nfsva, NULL, NULL); if (ret) { /* * Just set default values to get things going. */ NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr)); nfsva.na_vattr.va_type = VDIR; nfsva.na_vattr.va_mode = 0777; nfsva.na_vattr.va_nlink = 100; nfsva.na_vattr.va_uid = (uid_t)0; nfsva.na_vattr.va_gid = (gid_t)0; nfsva.na_vattr.va_fileid = 2; nfsva.na_vattr.va_gen = 1; nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE; nfsva.na_vattr.va_size = 512 * 1024; } } (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { mtx_lock(&nmp->nm_mtx); if (gotfsinfo || (nmp->nm_flag & NFSMNT_NFSV4)) nfscl_loadfsinfo(nmp, &fs); nfscl_loadsbinfo(nmp, &sb, sbp); sbp->f_iosize = newnfs_iosize(nmp); mtx_unlock(&nmp->nm_mtx); if (sbp != &mp->mnt_stat) { bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); } strncpy(&sbp->f_fstypename[0], mp->mnt_vfc->vfc_name, MFSNAMELEN); } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } vput(vp); vfs_unbusy(mp); return (error); } /* * nfs version 3 fsinfo rpc call */ int ncl_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred, struct thread *td) { struct nfsfsinfo fs; struct nfsvattr nfsva; int error, attrflag; error = nfsrpc_fsinfo(vp, &fs, cred, td, &nfsva, &attrflag, NULL); if (!error) { if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); mtx_lock(&nmp->nm_mtx); nfscl_loadfsinfo(nmp, &fs); mtx_unlock(&nmp->nm_mtx); } return (error); } /* * Mount a remote root fs via. nfs. This depends on the info in the * nfs_diskless structure that has been filled in properly by some primary * bootstrap. * It goes something like this: * - do enough of "ifconfig" by calling ifioctl() so that the system * can talk to the server * - If nfs_diskless.mygateway is filled in, use that address as * a default gateway. * - build the rootfs mount point and call mountnfs() to do the rest. * * It is assumed to be safe to read, modify, and write the nfsv3_diskless * structure, as well as other global NFS client variables here, as * nfs_mountroot() will be called once in the boot before any other NFS * client activity occurs. */ static int nfs_mountroot(struct mount *mp) { struct thread *td = curthread; struct nfsv3_diskless *nd = &nfsv3_diskless; struct socket *so; struct vnode *vp; struct ifreq ir; int error; u_long l; char buf[128]; char *cp; #if defined(BOOTP_NFSROOT) && defined(BOOTP) bootpc_init(); /* use bootp to get nfs_diskless filled in */ #elif defined(NFS_ROOT) nfs_setup_diskless(); #endif if (nfs_diskless_valid == 0) return (-1); if (nfs_diskless_valid == 1) nfs_convert_diskless(); /* * Do enough of ifconfig(8) so that the critical net interface can * talk to the server. */ error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0, td->td_ucred, td); if (error) panic("nfs_mountroot: socreate(%04x): %d", nd->myif.ifra_addr.sa_family, error); #if 0 /* XXX Bad idea */ /* * We might not have been told the right interface, so we pass * over the first ten interfaces of the same kind, until we get * one of them configured. */ for (i = strlen(nd->myif.ifra_name) - 1; nd->myif.ifra_name[i] >= '0' && nd->myif.ifra_name[i] <= '9'; nd->myif.ifra_name[i] ++) { error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); if(!error) break; } #endif error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); if (error) panic("nfs_mountroot: SIOCAIFADDR: %d", error); if ((cp = kern_getenv("boot.netif.mtu")) != NULL) { ir.ifr_mtu = strtol(cp, NULL, 10); bcopy(nd->myif.ifra_name, ir.ifr_name, IFNAMSIZ); freeenv(cp); error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td); if (error) printf("nfs_mountroot: SIOCSIFMTU: %d", error); } soclose(so); /* * If the gateway field is filled in, set it as the default route. * Note that pxeboot will set a default route of 0 if the route * is not set by the DHCP server. Check also for a value of 0 * to avoid panicking inappropriately in that situation. */ if (nd->mygateway.sin_len != 0 && nd->mygateway.sin_addr.s_addr != 0) { struct sockaddr_in mask, sin; bzero((caddr_t)&mask, sizeof(mask)); sin = mask; sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); /* XXX MRT use table 0 for this sort of thing */ CURVNET_SET(TD_TO_VNET(td)); error = rtrequest_fib(RTM_ADD, (struct sockaddr *)&sin, (struct sockaddr *)&nd->mygateway, (struct sockaddr *)&mask, RTF_UP | RTF_GATEWAY, NULL, RT_DEFAULT_FIB); CURVNET_RESTORE(); if (error) panic("nfs_mountroot: RTM_ADD: %d", error); } /* * Create the rootfs mount point. */ nd->root_args.fh = nd->root_fh; nd->root_args.fhsize = nd->root_fhsize; l = ntohl(nd->root_saddr.sin_addr.s_addr); snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", (l >> 24) & 0xff, (l >> 16) & 0xff, (l >> 8) & 0xff, (l >> 0) & 0xff, nd->root_hostnam); printf("NFS ROOT: %s\n", buf); nd->root_args.hostname = buf; if ((error = nfs_mountdiskless(buf, &nd->root_saddr, &nd->root_args, td, &vp, mp)) != 0) { return (error); } /* * This is not really an nfs issue, but it is much easier to * set hostname here and then let the "/etc/rc.xxx" files * mount the right /var based upon its preset value. */ mtx_lock(&prison0.pr_mtx); strlcpy(prison0.pr_hostname, nd->my_hostnam, sizeof(prison0.pr_hostname)); mtx_unlock(&prison0.pr_mtx); inittodr(ntohl(nd->root_time)); return (0); } /* * Internal version of mount system call for diskless setup. */ static int nfs_mountdiskless(char *path, struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, struct vnode **vpp, struct mount *mp) { struct sockaddr *nam; int dirlen, error; char *dirpath; /* * Find the directory path in "path", which also has the server's * name/ip address in it. */ dirpath = strchr(path, ':'); if (dirpath != NULL) dirlen = strlen(++dirpath); else dirlen = 0; nam = sodupsockaddr((struct sockaddr *)sin, M_WAITOK); if ((error = mountnfs(args, mp, nam, path, NULL, 0, dirpath, dirlen, NULL, 0, vpp, td->td_ucred, td, NFS_DEFAULT_NAMETIMEO, NFS_DEFAULT_NEGNAMETIMEO, 0)) != 0) { printf("nfs_mountroot: mount %s on /: %d\n", path, error); return (error); } return (0); } static void nfs_sec_name(char *sec, int *flagsp) { if (!strcmp(sec, "krb5")) *flagsp |= NFSMNT_KERB; else if (!strcmp(sec, "krb5i")) *flagsp |= (NFSMNT_KERB | NFSMNT_INTEGRITY); else if (!strcmp(sec, "krb5p")) *flagsp |= (NFSMNT_KERB | NFSMNT_PRIVACY); } static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp, const char *hostname, struct ucred *cred, struct thread *td) { int adjsock; char *p; /* * Set read-only flag if requested; otherwise, clear it if this is * an update. If this is not an update, then either the read-only * flag is already clear, or this is a root mount and it was set * intentionally at some previous point. */ if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) { MNT_ILOCK(mp); mp->mnt_flag |= MNT_RDONLY; MNT_IUNLOCK(mp); } else if (mp->mnt_flag & MNT_UPDATE) { MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_RDONLY; MNT_IUNLOCK(mp); } /* * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes * no sense in that context. Also, set up appropriate retransmit * and soft timeout behavior. */ if (argp->sotype == SOCK_STREAM) { nmp->nm_flag &= ~NFSMNT_NOCONN; nmp->nm_timeo = NFS_MAXTIMEO; if ((argp->flags & NFSMNT_NFSV4) != 0) nmp->nm_retry = INT_MAX; else nmp->nm_retry = NFS_RETRANS_TCP; } /* Also clear RDIRPLUS if NFSv2, it crashes some servers */ if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) { argp->flags &= ~NFSMNT_RDIRPLUS; nmp->nm_flag &= ~NFSMNT_RDIRPLUS; } /* Clear ONEOPENOWN for NFSv2, 3 and 4.0. */ if (nmp->nm_minorvers == 0) { argp->flags &= ~NFSMNT_ONEOPENOWN; nmp->nm_flag &= ~NFSMNT_ONEOPENOWN; } /* Re-bind if rsrvd port requested and wasn't on one */ adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) && (argp->flags & NFSMNT_RESVPORT); /* Also re-bind if we're switching to/from a connected UDP socket */ adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != (argp->flags & NFSMNT_NOCONN)); /* Update flags atomically. Don't change the lock bits. */ nmp->nm_flag = argp->flags | nmp->nm_flag; if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; if (nmp->nm_timeo < NFS_MINTIMEO) nmp->nm_timeo = NFS_MINTIMEO; else if (nmp->nm_timeo > NFS_MAXTIMEO) nmp->nm_timeo = NFS_MAXTIMEO; } if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { nmp->nm_retry = argp->retrans; if (nmp->nm_retry > NFS_MAXREXMIT) nmp->nm_retry = NFS_MAXREXMIT; } if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { nmp->nm_wsize = argp->wsize; /* * Clip at the power of 2 below the size. There is an * issue (not isolated) that causes intermittent page * faults if this is not done. */ if (nmp->nm_wsize > NFS_FABLKSIZE) nmp->nm_wsize = 1 << (fls(nmp->nm_wsize) - 1); else nmp->nm_wsize = NFS_FABLKSIZE; } if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { nmp->nm_rsize = argp->rsize; /* * Clip at the power of 2 below the size. There is an * issue (not isolated) that causes intermittent page * faults if this is not done. */ if (nmp->nm_rsize > NFS_FABLKSIZE) nmp->nm_rsize = 1 << (fls(nmp->nm_rsize) - 1); else nmp->nm_rsize = NFS_FABLKSIZE; } if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { nmp->nm_readdirsize = argp->readdirsize; } if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) nmp->nm_acregmin = argp->acregmin; else nmp->nm_acregmin = NFS_MINATTRTIMO; if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) nmp->nm_acregmax = argp->acregmax; else nmp->nm_acregmax = NFS_MAXATTRTIMO; if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) nmp->nm_acdirmin = argp->acdirmin; else nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) nmp->nm_acdirmax = argp->acdirmax; else nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; if (nmp->nm_acdirmin > nmp->nm_acdirmax) nmp->nm_acdirmin = nmp->nm_acdirmax; if (nmp->nm_acregmin > nmp->nm_acregmax) nmp->nm_acregmin = nmp->nm_acregmax; if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { if (argp->readahead <= NFS_MAXRAHEAD) nmp->nm_readahead = argp->readahead; else nmp->nm_readahead = NFS_MAXRAHEAD; } if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) { if (argp->wcommitsize < nmp->nm_wsize) nmp->nm_wcommitsize = nmp->nm_wsize; else nmp->nm_wcommitsize = argp->wcommitsize; } adjsock |= ((nmp->nm_sotype != argp->sotype) || (nmp->nm_soproto != argp->proto)); if (nmp->nm_client != NULL && adjsock) { int haslock = 0, error = 0; if (nmp->nm_sotype == SOCK_STREAM) { error = newnfs_sndlock(&nmp->nm_sockreq.nr_lock); if (!error) haslock = 1; } if (!error) { newnfs_disconnect(&nmp->nm_sockreq); if (haslock) newnfs_sndunlock(&nmp->nm_sockreq.nr_lock); nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; if (nmp->nm_sotype == SOCK_DGRAM) while (newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0)) { printf("newnfs_args: retrying connect\n"); (void) nfs_catnap(PSOCK, 0, "nfscon"); } } } else { nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; } if (hostname != NULL) { strlcpy(nmp->nm_hostname, hostname, sizeof(nmp->nm_hostname)); p = strchr(nmp->nm_hostname, ':'); if (p != NULL) *p = '\0'; } } static const char *nfs_opts[] = { "from", "nfs_args", "noac", "noatime", "noexec", "suiddir", "nosuid", "nosymfollow", "union", "noclusterr", "noclusterw", "multilabel", "acls", "force", "update", "async", "noconn", "nolockd", "conn", "lockd", "intr", "rdirplus", "readdirsize", "soft", "hard", "mntudp", "tcp", "udp", "wsize", "rsize", "retrans", "actimeo", "acregmin", "acregmax", "acdirmin", "acdirmax", "resvport", "readahead", "hostname", "timeo", "timeout", "addr", "fh", "nfsv3", "sec", "principal", "nfsv4", "gssname", "allgssname", "dirpath", "minorversion", "nametimeo", "negnametimeo", "nocto", "noncontigwr", "pnfs", "wcommitsize", "oneopenown", NULL }; /* * Parse the "from" mountarg, passed by the generic mount(8) program * or the mountroot code. This is used when rerooting into NFS. * * Note that the "hostname" is actually a "hostname:/share/path" string. */ static int nfs_mount_parse_from(struct vfsoptlist *opts, char **hostnamep, struct sockaddr_in **sinp, char *dirpath, size_t dirpathsize, int *dirlenp) { char *nam, *delimp, *hostp, *spec; int error, have_bracket = 0, offset, rv, speclen; struct sockaddr_in *sin; size_t len; error = vfs_getopt(opts, "from", (void **)&spec, &speclen); if (error != 0) return (error); nam = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK); /* * This part comes from sbin/mount_nfs/mount_nfs.c:getnfsargs(). */ if (*spec == '[' && (delimp = strchr(spec + 1, ']')) != NULL && *(delimp + 1) == ':') { hostp = spec + 1; spec = delimp + 2; have_bracket = 1; } else if ((delimp = strrchr(spec, ':')) != NULL) { hostp = spec; spec = delimp + 1; } else if ((delimp = strrchr(spec, '@')) != NULL) { printf("%s: path@server syntax is deprecated, " "use server:path\n", __func__); hostp = delimp + 1; } else { printf("%s: no : nfs-name\n", __func__); free(nam, M_TEMP); return (EINVAL); } *delimp = '\0'; /* * If there has been a trailing slash at mounttime it seems * that some mountd implementations fail to remove the mount * entries from their mountlist while unmounting. */ for (speclen = strlen(spec); speclen > 1 && spec[speclen - 1] == '/'; speclen--) spec[speclen - 1] = '\0'; if (strlen(hostp) + strlen(spec) + 1 > MNAMELEN) { printf("%s: %s:%s: name too long", __func__, hostp, spec); free(nam, M_TEMP); return (EINVAL); } /* Make both '@' and ':' notations equal */ if (*hostp != '\0') { len = strlen(hostp); offset = 0; if (have_bracket) nam[offset++] = '['; memmove(nam + offset, hostp, len); if (have_bracket) nam[len + offset++] = ']'; nam[len + offset++] = ':'; memmove(nam + len + offset, spec, speclen); nam[len + speclen + offset] = '\0'; } else nam[0] = '\0'; /* * XXX: IPv6 */ sin = malloc(sizeof(*sin), M_SONAME, M_WAITOK); rv = inet_pton(AF_INET, hostp, &sin->sin_addr); if (rv != 1) { printf("%s: cannot parse '%s', inet_pton() returned %d\n", __func__, hostp, rv); free(nam, M_TEMP); free(sin, M_SONAME); return (EINVAL); } sin->sin_len = sizeof(*sin); sin->sin_family = AF_INET; /* * XXX: hardcoded port number. */ sin->sin_port = htons(2049); *hostnamep = strdup(nam, M_NEWNFSMNT); *sinp = sin; strlcpy(dirpath, spec, dirpathsize); *dirlenp = strlen(dirpath); free(nam, M_TEMP); return (0); } /* * VFS Operations. * * mount system call * It seems a bit dumb to copyinstr() the host and path here and then * bcopy() them in mountnfs(), but I wanted to detect errors before * doing the getsockaddr() call because getsockaddr() allocates an mbuf and * an error after that means that I have to release the mbuf. */ /* ARGSUSED */ static int nfs_mount(struct mount *mp) { struct nfs_args args = { .version = NFS_ARGSVERSION, .addr = NULL, .addrlen = sizeof (struct sockaddr_in), .sotype = SOCK_STREAM, .proto = 0, .fh = NULL, .fhsize = 0, .flags = NFSMNT_RESVPORT, .wsize = NFS_WSIZE, .rsize = NFS_RSIZE, .readdirsize = NFS_READDIRSIZE, .timeo = 10, .retrans = NFS_RETRANS, .readahead = NFS_DEFRAHEAD, .wcommitsize = 0, /* was: NQ_DEFLEASE */ .hostname = NULL, .acregmin = NFS_MINATTRTIMO, .acregmax = NFS_MAXATTRTIMO, .acdirmin = NFS_MINDIRATTRTIMO, .acdirmax = NFS_MAXDIRATTRTIMO, }; int error = 0, ret, len; struct sockaddr *nam = NULL; struct vnode *vp; struct thread *td; char *hst; u_char nfh[NFSX_FHMAX], krbname[100], dirpath[100], srvkrbname[100]; char *cp, *opt, *name, *secname; int nametimeo = NFS_DEFAULT_NAMETIMEO; int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO; int minvers = 0; int dirlen, has_nfs_args_opt, has_nfs_from_opt, krbnamelen, srvkrbnamelen; size_t hstlen; has_nfs_args_opt = 0; has_nfs_from_opt = 0; hst = malloc(MNAMELEN, M_TEMP, M_WAITOK); if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) { error = EINVAL; goto out; } td = curthread; if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS && nfs_diskless_valid != 0) { error = nfs_mountroot(mp); goto out; } nfscl_init(); /* * The old mount_nfs program passed the struct nfs_args * from userspace to kernel. The new mount_nfs program * passes string options via nmount() from userspace to kernel * and we populate the struct nfs_args in the kernel. */ if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) { error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args, sizeof(args)); if (error != 0) goto out; if (args.version != NFS_ARGSVERSION) { error = EPROGMISMATCH; goto out; } has_nfs_args_opt = 1; } /* Handle the new style options. */ if (vfs_getopt(mp->mnt_optnew, "noac", NULL, NULL) == 0) { args.acdirmin = args.acdirmax = args.acregmin = args.acregmax = 0; args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX | NFSMNT_ACREGMIN | NFSMNT_ACREGMAX; } if (vfs_getopt(mp->mnt_optnew, "noconn", NULL, NULL) == 0) args.flags |= NFSMNT_NOCONN; if (vfs_getopt(mp->mnt_optnew, "conn", NULL, NULL) == 0) args.flags &= ~NFSMNT_NOCONN; if (vfs_getopt(mp->mnt_optnew, "nolockd", NULL, NULL) == 0) args.flags |= NFSMNT_NOLOCKD; if (vfs_getopt(mp->mnt_optnew, "lockd", NULL, NULL) == 0) args.flags &= ~NFSMNT_NOLOCKD; if (vfs_getopt(mp->mnt_optnew, "intr", NULL, NULL) == 0) args.flags |= NFSMNT_INT; if (vfs_getopt(mp->mnt_optnew, "rdirplus", NULL, NULL) == 0) args.flags |= NFSMNT_RDIRPLUS; if (vfs_getopt(mp->mnt_optnew, "resvport", NULL, NULL) == 0) args.flags |= NFSMNT_RESVPORT; if (vfs_getopt(mp->mnt_optnew, "noresvport", NULL, NULL) == 0) args.flags &= ~NFSMNT_RESVPORT; if (vfs_getopt(mp->mnt_optnew, "soft", NULL, NULL) == 0) args.flags |= NFSMNT_SOFT; if (vfs_getopt(mp->mnt_optnew, "hard", NULL, NULL) == 0) args.flags &= ~NFSMNT_SOFT; if (vfs_getopt(mp->mnt_optnew, "mntudp", NULL, NULL) == 0) args.sotype = SOCK_DGRAM; if (vfs_getopt(mp->mnt_optnew, "udp", NULL, NULL) == 0) args.sotype = SOCK_DGRAM; if (vfs_getopt(mp->mnt_optnew, "tcp", NULL, NULL) == 0) args.sotype = SOCK_STREAM; if (vfs_getopt(mp->mnt_optnew, "nfsv3", NULL, NULL) == 0) args.flags |= NFSMNT_NFSV3; if (vfs_getopt(mp->mnt_optnew, "nfsv4", NULL, NULL) == 0) { args.flags |= NFSMNT_NFSV4; args.sotype = SOCK_STREAM; } if (vfs_getopt(mp->mnt_optnew, "allgssname", NULL, NULL) == 0) args.flags |= NFSMNT_ALLGSSNAME; if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0) args.flags |= NFSMNT_NOCTO; if (vfs_getopt(mp->mnt_optnew, "noncontigwr", NULL, NULL) == 0) args.flags |= NFSMNT_NONCONTIGWR; if (vfs_getopt(mp->mnt_optnew, "pnfs", NULL, NULL) == 0) args.flags |= NFSMNT_PNFS; if (vfs_getopt(mp->mnt_optnew, "oneopenown", NULL, NULL) == 0) args.flags |= NFSMNT_ONEOPENOWN; if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal readdirsize"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.readdirsize); if (ret != 1 || args.readdirsize <= 0) { vfs_mount_error(mp, "illegal readdirsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_READDIRSIZE; } if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal readahead"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.readahead); if (ret != 1 || args.readahead <= 0) { vfs_mount_error(mp, "illegal readahead: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_READAHEAD; } if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal wsize"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.wsize); if (ret != 1 || args.wsize <= 0) { vfs_mount_error(mp, "illegal wsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_WSIZE; } if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal rsize"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.rsize); if (ret != 1 || args.rsize <= 0) { vfs_mount_error(mp, "illegal wsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_RSIZE; } if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) { if (opt == NULL) { vfs_mount_error(mp, "illegal retrans"); error = EINVAL; goto out; } ret = sscanf(opt, "%d", &args.retrans); if (ret != 1 || args.retrans <= 0) { vfs_mount_error(mp, "illegal retrans: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_RETRANS; } if (vfs_getopt(mp->mnt_optnew, "actimeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acregmin); if (ret != 1 || args.acregmin < 0) { vfs_mount_error(mp, "illegal actimeo: %s", opt); error = EINVAL; goto out; } args.acdirmin = args.acdirmax = args.acregmax = args.acregmin; args.flags |= NFSMNT_ACDIRMIN | NFSMNT_ACDIRMAX | NFSMNT_ACREGMIN | NFSMNT_ACREGMAX; } if (vfs_getopt(mp->mnt_optnew, "acregmin", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acregmin); if (ret != 1 || args.acregmin < 0) { vfs_mount_error(mp, "illegal acregmin: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACREGMIN; } if (vfs_getopt(mp->mnt_optnew, "acregmax", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acregmax); if (ret != 1 || args.acregmax < 0) { vfs_mount_error(mp, "illegal acregmax: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACREGMAX; } if (vfs_getopt(mp->mnt_optnew, "acdirmin", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acdirmin); if (ret != 1 || args.acdirmin < 0) { vfs_mount_error(mp, "illegal acdirmin: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACDIRMIN; } if (vfs_getopt(mp->mnt_optnew, "acdirmax", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.acdirmax); if (ret != 1 || args.acdirmax < 0) { vfs_mount_error(mp, "illegal acdirmax: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_ACDIRMAX; } if (vfs_getopt(mp->mnt_optnew, "wcommitsize", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.wcommitsize); if (ret != 1 || args.wcommitsize < 0) { vfs_mount_error(mp, "illegal wcommitsize: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_WCOMMITSIZE; } if (vfs_getopt(mp->mnt_optnew, "timeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.timeo); if (ret != 1 || args.timeo <= 0) { vfs_mount_error(mp, "illegal timeo: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_TIMEO; } if (vfs_getopt(mp->mnt_optnew, "timeout", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &args.timeo); if (ret != 1 || args.timeo <= 0) { vfs_mount_error(mp, "illegal timeout: %s", opt); error = EINVAL; goto out; } args.flags |= NFSMNT_TIMEO; } if (vfs_getopt(mp->mnt_optnew, "nametimeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &nametimeo); if (ret != 1 || nametimeo < 0) { vfs_mount_error(mp, "illegal nametimeo: %s", opt); error = EINVAL; goto out; } } if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &negnametimeo); if (ret != 1 || negnametimeo < 0) { vfs_mount_error(mp, "illegal negnametimeo: %s", opt); error = EINVAL; goto out; } } if (vfs_getopt(mp->mnt_optnew, "minorversion", (void **)&opt, NULL) == 0) { ret = sscanf(opt, "%d", &minvers); if (ret != 1 || minvers < 0 || minvers > 1 || (args.flags & NFSMNT_NFSV4) == 0) { vfs_mount_error(mp, "illegal minorversion: %s", opt); error = EINVAL; goto out; } } if (vfs_getopt(mp->mnt_optnew, "sec", (void **) &secname, NULL) == 0) nfs_sec_name(secname, &args.flags); if (mp->mnt_flag & MNT_UPDATE) { struct nfsmount *nmp = VFSTONFS(mp); if (nmp == NULL) { error = EIO; goto out; } /* * If a change from TCP->UDP is done and there are thread(s) * that have I/O RPC(s) in progress with a transfer size * greater than NFS_MAXDGRAMDATA, those thread(s) will be * hung, retrying the RPC(s) forever. Usually these threads * will be seen doing an uninterruptible sleep on wait channel * "nfsreq". */ if (args.sotype == SOCK_DGRAM && nmp->nm_sotype == SOCK_STREAM) tprintf(td->td_proc, LOG_WARNING, "Warning: mount -u that changes TCP->UDP can result in hung threads\n"); /* * When doing an update, we can't change version, * security, switch lockd strategies, change cookie * translation or switch oneopenown. */ args.flags = (args.flags & ~(NFSMNT_NFSV3 | NFSMNT_NFSV4 | NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY | NFSMNT_ONEOPENOWN | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) | (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4 | NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY | NFSMNT_ONEOPENOWN | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)); nfs_decode_args(mp, nmp, &args, NULL, td->td_ucred, td); goto out; } /* * Make the nfs_ip_paranoia sysctl serve as the default connection * or no-connection mode for those protocols that support * no-connection mode (the flag will be cleared later for protocols * that do not support no-connection mode). This will allow a client * to receive replies from a different IP then the request was * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), * not 0. */ if (nfs_ip_paranoia == 0) args.flags |= NFSMNT_NOCONN; if (has_nfs_args_opt != 0) { /* * In the 'nfs_args' case, the pointers in the args * structure are in userland - we copy them in here. */ if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) { vfs_mount_error(mp, "Bad file handle"); error = EINVAL; goto out; } error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); if (error != 0) goto out; error = copyinstr(args.hostname, hst, MNAMELEN - 1, &hstlen); if (error != 0) goto out; bzero(&hst[hstlen], MNAMELEN - hstlen); args.hostname = hst; /* getsockaddr() call must be after above copyin() calls */ error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); if (error != 0) goto out; } else if (nfs_mount_parse_from(mp->mnt_optnew, &args.hostname, (struct sockaddr_in **)&nam, dirpath, sizeof(dirpath), &dirlen) == 0) { has_nfs_from_opt = 1; bcopy(args.hostname, hst, MNAMELEN); hst[MNAMELEN - 1] = '\0'; /* * This only works with NFSv4 for now. */ args.fhsize = 0; args.flags |= NFSMNT_NFSV4; args.sotype = SOCK_STREAM; } else { if (vfs_getopt(mp->mnt_optnew, "fh", (void **)&args.fh, &args.fhsize) == 0) { if (args.fhsize < 0 || args.fhsize > NFSX_FHMAX) { vfs_mount_error(mp, "Bad file handle"); error = EINVAL; goto out; } bcopy(args.fh, nfh, args.fhsize); } else { args.fhsize = 0; } (void) vfs_getopt(mp->mnt_optnew, "hostname", (void **)&args.hostname, &len); if (args.hostname == NULL) { vfs_mount_error(mp, "Invalid hostname"); error = EINVAL; goto out; } if (len >= MNAMELEN) { vfs_mount_error(mp, "Hostname too long"); error = EINVAL; goto out; } bcopy(args.hostname, hst, len); hst[len] = '\0'; } if (vfs_getopt(mp->mnt_optnew, "principal", (void **)&name, NULL) == 0) strlcpy(srvkrbname, name, sizeof (srvkrbname)); else { snprintf(srvkrbname, sizeof (srvkrbname), "nfs@%s", hst); cp = strchr(srvkrbname, ':'); if (cp != NULL) *cp = '\0'; } srvkrbnamelen = strlen(srvkrbname); if (vfs_getopt(mp->mnt_optnew, "gssname", (void **)&name, NULL) == 0) strlcpy(krbname, name, sizeof (krbname)); else krbname[0] = '\0'; krbnamelen = strlen(krbname); if (has_nfs_from_opt == 0) { if (vfs_getopt(mp->mnt_optnew, "dirpath", (void **)&name, NULL) == 0) strlcpy(dirpath, name, sizeof (dirpath)); else dirpath[0] = '\0'; dirlen = strlen(dirpath); } if (has_nfs_args_opt == 0 && has_nfs_from_opt == 0) { if (vfs_getopt(mp->mnt_optnew, "addr", (void **)&args.addr, &args.addrlen) == 0) { if (args.addrlen > SOCK_MAXADDRLEN) { error = ENAMETOOLONG; goto out; } nam = malloc(args.addrlen, M_SONAME, M_WAITOK); bcopy(args.addr, nam, args.addrlen); nam->sa_len = args.addrlen; } else { vfs_mount_error(mp, "No server address"); error = EINVAL; goto out; } } args.fh = nfh; error = mountnfs(&args, mp, nam, hst, krbname, krbnamelen, dirpath, dirlen, srvkrbname, srvkrbnamelen, &vp, td->td_ucred, td, nametimeo, negnametimeo, minvers); out: if (!error) { MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_NO_IOPF | MNTK_USES_BCACHE; if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV4) != 0) mp->mnt_kern_flag |= MNTK_NULL_NOCACHE; MNT_IUNLOCK(mp); } free(hst, M_TEMP); return (error); } /* * VFS Operations. * * mount system call * It seems a bit dumb to copyinstr() the host and path here and then * bcopy() them in mountnfs(), but I wanted to detect errors before * doing the getsockaddr() call because getsockaddr() allocates an mbuf and * an error after that means that I have to release the mbuf. */ /* ARGSUSED */ static int nfs_cmount(struct mntarg *ma, void *data, uint64_t flags) { int error; struct nfs_args args; error = copyin(data, &args, sizeof (struct nfs_args)); if (error) return error; ma = mount_arg(ma, "nfs_args", &args, sizeof args); error = kernel_mount(ma, flags); return (error); } /* * Common code for mount and mountroot */ static int mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, char *hst, u_char *krbname, int krbnamelen, u_char *dirpath, int dirlen, u_char *srvkrbname, int srvkrbnamelen, struct vnode **vpp, struct ucred *cred, struct thread *td, int nametimeo, int negnametimeo, int minvers) { struct nfsmount *nmp; struct nfsnode *np; int error, trycnt, ret; struct nfsvattr nfsva; struct nfsclclient *clp; struct nfsclds *dsp, *tdsp; uint32_t lease; static u_int64_t clval = 0; NFSCL_DEBUG(3, "in mnt\n"); clp = NULL; if (mp->mnt_flag & MNT_UPDATE) { nmp = VFSTONFS(mp); printf("%s: MNT_UPDATE is no longer handled here\n", __func__); FREE(nam, M_SONAME); return (0); } else { MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount) + krbnamelen + dirlen + srvkrbnamelen + 2, M_NEWNFSMNT, M_WAITOK | M_ZERO); TAILQ_INIT(&nmp->nm_bufq); TAILQ_INIT(&nmp->nm_sess); if (clval == 0) clval = (u_int64_t)nfsboottime.tv_sec; nmp->nm_clval = clval++; nmp->nm_krbnamelen = krbnamelen; nmp->nm_dirpathlen = dirlen; nmp->nm_srvkrbnamelen = srvkrbnamelen; if (td->td_ucred->cr_uid != (uid_t)0) { /* * nm_uid is used to get KerberosV credentials for * the nfsv4 state handling operations if there is * no host based principal set. Use the uid of * this user if not root, since they are doing the * mount. I don't think setting this for root will * work, since root normally does not have user * credentials in a credentials cache. */ nmp->nm_uid = td->td_ucred->cr_uid; } else { /* * Just set to -1, so it won't be used. */ nmp->nm_uid = (uid_t)-1; } /* Copy and null terminate all the names */ if (nmp->nm_krbnamelen > 0) { bcopy(krbname, nmp->nm_krbname, nmp->nm_krbnamelen); nmp->nm_name[nmp->nm_krbnamelen] = '\0'; } if (nmp->nm_dirpathlen > 0) { bcopy(dirpath, NFSMNT_DIRPATH(nmp), nmp->nm_dirpathlen); nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen + 1] = '\0'; } if (nmp->nm_srvkrbnamelen > 0) { bcopy(srvkrbname, NFSMNT_SRVKRBNAME(nmp), nmp->nm_srvkrbnamelen); nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen + nmp->nm_srvkrbnamelen + 2] = '\0'; } nmp->nm_sockreq.nr_cred = crhold(cred); mtx_init(&nmp->nm_sockreq.nr_mtx, "nfssock", NULL, MTX_DEF); mp->mnt_data = nmp; nmp->nm_getinfo = nfs_getnlminfo; nmp->nm_vinvalbuf = ncl_vinvalbuf; } vfs_getnewfsid(mp); nmp->nm_mountp = mp; mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF | MTX_DUPOK); /* * Since nfs_decode_args() might optionally set them, these * need to be set to defaults before the call, so that the * optional settings aren't overwritten. */ nmp->nm_nametimeo = nametimeo; nmp->nm_negnametimeo = negnametimeo; nmp->nm_timeo = NFS_TIMEO; nmp->nm_retry = NFS_RETRANS; nmp->nm_readahead = NFS_DEFRAHEAD; /* This is empirical approximation of sqrt(hibufspace) * 256. */ nmp->nm_wcommitsize = NFS_MAXBSIZE / 256; while ((long)nmp->nm_wcommitsize * nmp->nm_wcommitsize < hibufspace) nmp->nm_wcommitsize *= 2; nmp->nm_wcommitsize *= 256; if ((argp->flags & NFSMNT_NFSV4) != 0) nmp->nm_minorvers = minvers; else nmp->nm_minorvers = 0; nfs_decode_args(mp, nmp, argp, hst, cred, td); /* * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too * high, depending on whether we end up with negative offsets in * the client or server somewhere. 2GB-1 may be safer. * * For V3, ncl_fsinfo will adjust this as necessary. Assume maximum * that we can handle until we find out otherwise. */ if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) nmp->nm_maxfilesize = 0xffffffffLL; else nmp->nm_maxfilesize = OFF_MAX; if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) { nmp->nm_wsize = NFS_WSIZE; nmp->nm_rsize = NFS_RSIZE; nmp->nm_readdirsize = NFS_READDIRSIZE; } nmp->nm_numgrps = NFS_MAXGRPS; nmp->nm_tprintf_delay = nfs_tprintf_delay; if (nmp->nm_tprintf_delay < 0) nmp->nm_tprintf_delay = 0; nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay; if (nmp->nm_tprintf_initial_delay < 0) nmp->nm_tprintf_initial_delay = 0; nmp->nm_fhsize = argp->fhsize; if (nmp->nm_fhsize > 0) bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); nmp->nm_nam = nam; /* Set up the sockets and per-host congestion */ nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; nmp->nm_sockreq.nr_prog = NFS_PROG; if ((argp->flags & NFSMNT_NFSV4)) nmp->nm_sockreq.nr_vers = NFS_VER4; else if ((argp->flags & NFSMNT_NFSV3)) nmp->nm_sockreq.nr_vers = NFS_VER3; else nmp->nm_sockreq.nr_vers = NFS_VER2; if ((error = newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0))) goto bad; /* For NFSv4.1, get the clientid now. */ if (nmp->nm_minorvers > 0) { NFSCL_DEBUG(3, "at getcl\n"); error = nfscl_getcl(mp, cred, td, 0, &clp); NFSCL_DEBUG(3, "aft getcl=%d\n", error); if (error != 0) goto bad; } if (nmp->nm_fhsize == 0 && (nmp->nm_flag & NFSMNT_NFSV4) && nmp->nm_dirpathlen > 0) { NFSCL_DEBUG(3, "in dirp\n"); /* * If the fhsize on the mount point == 0 for V4, the mount * path needs to be looked up. */ trycnt = 3; do { error = nfsrpc_getdirpath(nmp, NFSMNT_DIRPATH(nmp), cred, td); NFSCL_DEBUG(3, "aft dirp=%d\n", error); if (error) (void) nfs_catnap(PZERO, error, "nfsgetdirp"); } while (error && --trycnt > 0); if (error) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); goto bad; } } /* * A reference count is needed on the nfsnode representing the * remote root. If this object is not persistent, then backward * traversals of the mount point (i.e. "..") will not work if * the nfsnode gets flushed out of the cache. Ufs does not have * this problem, because one can identify root inodes by their * number == UFS_ROOTINO (2). */ if (nmp->nm_fhsize > 0) { /* * Set f_iosize to NFS_DIRBLKSIZ so that bo_bsize gets set * non-zero for the root vnode. f_iosize will be set correctly * by nfs_statfs() before any I/O occurs. */ mp->mnt_stat.f_iosize = NFS_DIRBLKSIZ; error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); if (error) goto bad; *vpp = NFSTOV(np); /* * Get file attributes and transfer parameters for the * mountpoint. This has the side effect of filling in * (*vpp)->v_type with the correct value. */ ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1, cred, td, &nfsva, NULL, &lease); if (ret) { /* * Just set default values to get things going. */ NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr)); nfsva.na_vattr.va_type = VDIR; nfsva.na_vattr.va_mode = 0777; nfsva.na_vattr.va_nlink = 100; nfsva.na_vattr.va_uid = (uid_t)0; nfsva.na_vattr.va_gid = (gid_t)0; nfsva.na_vattr.va_fileid = 2; nfsva.na_vattr.va_gen = 1; nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE; nfsva.na_vattr.va_size = 512 * 1024; lease = 60; } (void) nfscl_loadattrcache(vpp, &nfsva, NULL, NULL, 0, 1); if (nmp->nm_minorvers > 0) { NFSCL_DEBUG(3, "lease=%d\n", (int)lease); NFSLOCKCLSTATE(); clp->nfsc_renew = NFSCL_RENEW(lease); clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew; clp->nfsc_clientidrev++; if (clp->nfsc_clientidrev == 0) clp->nfsc_clientidrev++; NFSUNLOCKCLSTATE(); /* * Mount will succeed, so the renew thread can be * started now. */ nfscl_start_renewthread(clp); nfscl_clientrelease(clp); } if (argp->flags & NFSMNT_NFSV3) ncl_fsinfo(nmp, *vpp, cred, td); /* Mark if the mount point supports NFSv4 ACLs. */ if ((argp->flags & NFSMNT_NFSV4) != 0 && nfsrv_useacl != 0 && ret == 0 && NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) { MNT_ILOCK(mp); mp->mnt_flag |= MNT_NFS4ACLS; MNT_IUNLOCK(mp); } /* * Lose the lock but keep the ref. */ NFSVOPUNLOCK(*vpp, 0); return (0); } error = EIO; bad: if (clp != NULL) nfscl_clientrelease(clp); newnfs_disconnect(&nmp->nm_sockreq); crfree(nmp->nm_sockreq.nr_cred); if (nmp->nm_sockreq.nr_auth != NULL) AUTH_DESTROY(nmp->nm_sockreq.nr_auth); mtx_destroy(&nmp->nm_sockreq.nr_mtx); mtx_destroy(&nmp->nm_mtx); if (nmp->nm_clp != NULL) { NFSLOCKCLSTATE(); LIST_REMOVE(nmp->nm_clp, nfsc_list); NFSUNLOCKCLSTATE(); free(nmp->nm_clp, M_NFSCLCLIENT); } TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp) { if (dsp != TAILQ_FIRST(&nmp->nm_sess) && dsp->nfsclds_sockp != NULL) newnfs_disconnect(dsp->nfsclds_sockp); nfscl_freenfsclds(dsp); } FREE(nmp, M_NEWNFSMNT); FREE(nam, M_SONAME); return (error); } /* * unmount system call */ static int nfs_unmount(struct mount *mp, int mntflags) { struct thread *td; struct nfsmount *nmp; int error, flags = 0, i, trycnt = 0; struct nfsclds *dsp, *tdsp; td = curthread; if (mntflags & MNT_FORCE) flags |= FORCECLOSE; nmp = VFSTONFS(mp); /* * Goes something like this.. * - Call vflush() to clear out vnodes for this filesystem * - Close the socket * - Free up the data structures */ /* In the forced case, cancel any outstanding requests. */ if (mntflags & MNT_FORCE) { error = newnfs_nmcancelreqs(nmp); if (error) goto out; /* For a forced close, get rid of the renew thread now */ nfscl_umount(nmp, td); } /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ do { error = vflush(mp, 1, flags, td); if ((mntflags & MNT_FORCE) && error != 0 && ++trycnt < 30) (void) nfs_catnap(PSOCK, error, "newndm"); } while ((mntflags & MNT_FORCE) && error != 0 && trycnt < 30); if (error) goto out; /* * We are now committed to the unmount. */ if ((mntflags & MNT_FORCE) == 0) nfscl_umount(nmp, td); /* Make sure no nfsiods are assigned to this mount. */ mtx_lock(&ncl_iod_mutex); for (i = 0; i < NFS_MAXASYNCDAEMON; i++) if (ncl_iodmount[i] == nmp) { ncl_iodwant[i] = NFSIOD_AVAILABLE; ncl_iodmount[i] = NULL; } mtx_unlock(&ncl_iod_mutex); newnfs_disconnect(&nmp->nm_sockreq); crfree(nmp->nm_sockreq.nr_cred); FREE(nmp->nm_nam, M_SONAME); if (nmp->nm_sockreq.nr_auth != NULL) AUTH_DESTROY(nmp->nm_sockreq.nr_auth); mtx_destroy(&nmp->nm_sockreq.nr_mtx); mtx_destroy(&nmp->nm_mtx); TAILQ_FOREACH_SAFE(dsp, &nmp->nm_sess, nfsclds_list, tdsp) { if (dsp != TAILQ_FIRST(&nmp->nm_sess) && dsp->nfsclds_sockp != NULL) newnfs_disconnect(dsp->nfsclds_sockp); nfscl_freenfsclds(dsp); } FREE(nmp, M_NEWNFSMNT); out: return (error); } /* * Return root of a filesystem */ static int nfs_root(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; struct nfsmount *nmp; struct nfsnode *np; int error; nmp = VFSTONFS(mp); error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, flags); if (error) return error; vp = NFSTOV(np); /* * Get transfer parameters and attributes for root vnode once. */ mtx_lock(&nmp->nm_mtx); if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) { mtx_unlock(&nmp->nm_mtx); ncl_fsinfo(nmp, vp, curthread->td_ucred, curthread); } else mtx_unlock(&nmp->nm_mtx); if (vp->v_type == VNON) vp->v_type = VDIR; vp->v_vflag |= VV_ROOT; *vpp = vp; return (0); } /* * Flush out the buffer cache */ /* ARGSUSED */ static int nfs_sync(struct mount *mp, int waitfor) { struct vnode *vp, *mvp; struct thread *td; int error, allerror = 0; td = curthread; MNT_ILOCK(mp); /* * If a forced dismount is in progress, return from here so that * the umount(2) syscall doesn't get stuck in VFS_SYNC() before * calling VFS_UNMOUNT(). */ - if ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) { + if (NFSCL_FORCEDISM(mp)) { MNT_IUNLOCK(mp); return (EBADF); } MNT_IUNLOCK(mp); /* * Force stale buffer cache information to be flushed. */ loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* XXX Racy bv_cnt check. */ if (NFSVOPISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 || waitfor == MNT_LAZY) { VI_UNLOCK(vp); continue; } if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } error = VOP_FSYNC(vp, waitfor, td); if (error) allerror = error; NFSVOPUNLOCK(vp, 0); vrele(vp); } return (allerror); } static int nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req) { struct nfsmount *nmp = VFSTONFS(mp); struct vfsquery vq; int error; bzero(&vq, sizeof(vq)); switch (op) { #if 0 case VFS_CTL_NOLOCKS: val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0; if (req->oldptr != NULL) { error = SYSCTL_OUT(req, &val, sizeof(val)); if (error) return (error); } if (req->newptr != NULL) { error = SYSCTL_IN(req, &val, sizeof(val)); if (error) return (error); if (val) nmp->nm_flag |= NFSMNT_NOLOCKS; else nmp->nm_flag &= ~NFSMNT_NOLOCKS; } break; #endif case VFS_CTL_QUERY: mtx_lock(&nmp->nm_mtx); if (nmp->nm_state & NFSSTA_TIMEO) vq.vq_flags |= VQ_NOTRESP; mtx_unlock(&nmp->nm_mtx); #if 0 if (!(nmp->nm_flag & NFSMNT_NOLOCKS) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) vq.vq_flags |= VQ_NOTRESPLOCK; #endif error = SYSCTL_OUT(req, &vq, sizeof(vq)); break; case VFS_CTL_TIMEO: if (req->oldptr != NULL) { error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay, sizeof(nmp->nm_tprintf_initial_delay)); if (error) return (error); } if (req->newptr != NULL) { error = vfs_suser(mp, req->td); if (error) return (error); error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay, sizeof(nmp->nm_tprintf_initial_delay)); if (error) return (error); if (nmp->nm_tprintf_initial_delay < 0) nmp->nm_tprintf_initial_delay = 0; } break; default: return (ENOTSUP); } return (0); } /* * Purge any RPCs in progress, so that they will all return errors. * This allows dounmount() to continue as far as VFS_UNMOUNT() for a * forced dismount. */ static void nfs_purge(struct mount *mp) { struct nfsmount *nmp = VFSTONFS(mp); newnfs_nmcancelreqs(nmp); } /* * Extract the information needed by the nlm from the nfs vnode. */ static void nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp, struct sockaddr_storage *sp, int *is_v3p, off_t *sizep, struct timeval *timeop) { struct nfsmount *nmp; struct nfsnode *np = VTONFS(vp); nmp = VFSTONFS(vp->v_mount); if (fhlenp != NULL) *fhlenp = (size_t)np->n_fhp->nfh_len; if (fhp != NULL) bcopy(np->n_fhp->nfh_fh, fhp, np->n_fhp->nfh_len); if (sp != NULL) bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp))); if (is_v3p != NULL) *is_v3p = NFS_ISV3(vp); if (sizep != NULL) *sizep = np->n_size; if (timeop != NULL) { timeop->tv_sec = nmp->nm_timeo / NFS_HZ; timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ); } } /* * This function prints out an option name, based on the conditional * argument. */ static __inline void nfscl_printopt(struct nfsmount *nmp, int testval, char *opt, char **buf, size_t *blen) { int len; if (testval != 0 && *blen > strlen(opt)) { len = snprintf(*buf, *blen, "%s", opt); if (len != strlen(opt)) printf("EEK!!\n"); *buf += len; *blen -= len; } } /* * This function printf out an options integer value. */ static __inline void nfscl_printoptval(struct nfsmount *nmp, int optval, char *opt, char **buf, size_t *blen) { int len; if (*blen > strlen(opt) + 1) { /* Could result in truncated output string. */ len = snprintf(*buf, *blen, "%s=%d", opt, optval); if (len < *blen) { *buf += len; *blen -= len; } } } /* * Load the option flags and values into the buffer. */ void nfscl_retopts(struct nfsmount *nmp, char *buffer, size_t buflen) { char *buf; size_t blen; buf = buffer; blen = buflen; nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV4) != 0, "nfsv4", &buf, &blen); if ((nmp->nm_flag & NFSMNT_NFSV4) != 0) { nfscl_printoptval(nmp, nmp->nm_minorvers, ",minorversion", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_PNFS) != 0, ",pnfs", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_ONEOPENOWN) != 0 && nmp->nm_minorvers > 0, ",oneopenown", &buf, &blen); } nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NFSV3) != 0, "nfsv3", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0, "nfsv2", &buf, &blen); nfscl_printopt(nmp, nmp->nm_sotype == SOCK_STREAM, ",tcp", &buf, &blen); nfscl_printopt(nmp, nmp->nm_sotype != SOCK_STREAM, ",udp", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RESVPORT) != 0, ",resvport", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCONN) != 0, ",noconn", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) == 0, ",hard", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_SOFT) != 0, ",soft", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_INT) != 0, ",intr", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) == 0, ",cto", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) != 0, ",nocto", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0, ",noncontigwr", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) == 0, ",lockd", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) == NFSMNT_NOLOCKD, ",nolockd", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_RDIRPLUS) != 0, ",rdirplus", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_KERB) == 0, ",sec=sys", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY)) == NFSMNT_KERB, ",sec=krb5", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_INTEGRITY), ",sec=krb5i", &buf, &blen); nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_KERB | NFSMNT_INTEGRITY | NFSMNT_PRIVACY)) == (NFSMNT_KERB | NFSMNT_PRIVACY), ",sec=krb5p", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acdirmin, ",acdirmin", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acdirmax, ",acdirmax", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acregmin, ",acregmin", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_acregmax, ",acregmax", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_nametimeo, ",nametimeo", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_negnametimeo, ",negnametimeo", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_rsize, ",rsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_wsize, ",wsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_readdirsize, ",readdirsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_readahead, ",readahead", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_wcommitsize, ",wcommitsize", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_timeo, ",timeout", &buf, &blen); nfscl_printoptval(nmp, nmp->nm_retry, ",retrans", &buf, &blen); } Index: head/sys/fs/nfsclient/nfs_clvnops.c =================================================================== --- head/sys/fs/nfsclient/nfs_clvnops.c (revision 321627) +++ head/sys/fs/nfsclient/nfs_clvnops.c (revision 321628) @@ -1,3546 +1,3546 @@ /*- * 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. * * from nfs_vnops.c 8.16 (Berkeley) 5/27/95 */ #include __FBSDID("$FreeBSD$"); /* * vnode op calls for Sun NFS version 2, 3 and 4 */ #include "opt_inet.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 #include #include #include #include #include #include #include #ifdef KDTRACE_HOOKS #include dtrace_nfsclient_accesscache_flush_probe_func_t dtrace_nfscl_accesscache_flush_done_probe; uint32_t nfscl_accesscache_flush_done_id; dtrace_nfsclient_accesscache_get_probe_func_t dtrace_nfscl_accesscache_get_hit_probe, dtrace_nfscl_accesscache_get_miss_probe; uint32_t nfscl_accesscache_get_hit_id; uint32_t nfscl_accesscache_get_miss_id; dtrace_nfsclient_accesscache_load_probe_func_t dtrace_nfscl_accesscache_load_done_probe; uint32_t nfscl_accesscache_load_done_id; #endif /* !KDTRACE_HOOKS */ /* Defs */ #define TRUE 1 #define FALSE 0 extern struct nfsstatsv1 nfsstatsv1; extern int nfsrv_useacl; extern int nfscl_debuglevel; MALLOC_DECLARE(M_NEWNFSREQ); static vop_read_t nfsfifo_read; static vop_write_t nfsfifo_write; static vop_close_t nfsfifo_close; static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *, struct thread *); static vop_lookup_t nfs_lookup; static vop_create_t nfs_create; static vop_mknod_t nfs_mknod; static vop_open_t nfs_open; static vop_pathconf_t nfs_pathconf; static vop_close_t nfs_close; static vop_access_t nfs_access; static vop_getattr_t nfs_getattr; static vop_setattr_t nfs_setattr; static vop_read_t nfs_read; static vop_fsync_t nfs_fsync; static vop_remove_t nfs_remove; static vop_link_t nfs_link; static vop_rename_t nfs_rename; static vop_mkdir_t nfs_mkdir; static vop_rmdir_t nfs_rmdir; static vop_symlink_t nfs_symlink; static vop_readdir_t nfs_readdir; static vop_strategy_t nfs_strategy; static int nfs_lookitup(struct vnode *, char *, int, struct ucred *, struct thread *, struct nfsnode **); static int nfs_sillyrename(struct vnode *, struct vnode *, struct componentname *); static vop_access_t nfsspec_access; static vop_readlink_t nfs_readlink; static vop_print_t nfs_print; static vop_advlock_t nfs_advlock; static vop_advlockasync_t nfs_advlockasync; static vop_getacl_t nfs_getacl; static vop_setacl_t nfs_setacl; static vop_set_text_t nfs_set_text; /* * Global vfs data structures for nfs */ struct vop_vector newnfs_vnodeops = { .vop_default = &default_vnodeops, .vop_access = nfs_access, .vop_advlock = nfs_advlock, .vop_advlockasync = nfs_advlockasync, .vop_close = nfs_close, .vop_create = nfs_create, .vop_fsync = nfs_fsync, .vop_getattr = nfs_getattr, .vop_getpages = ncl_getpages, .vop_putpages = ncl_putpages, .vop_inactive = ncl_inactive, .vop_link = nfs_link, .vop_lookup = nfs_lookup, .vop_mkdir = nfs_mkdir, .vop_mknod = nfs_mknod, .vop_open = nfs_open, .vop_pathconf = nfs_pathconf, .vop_print = nfs_print, .vop_read = nfs_read, .vop_readdir = nfs_readdir, .vop_readlink = nfs_readlink, .vop_reclaim = ncl_reclaim, .vop_remove = nfs_remove, .vop_rename = nfs_rename, .vop_rmdir = nfs_rmdir, .vop_setattr = nfs_setattr, .vop_strategy = nfs_strategy, .vop_symlink = nfs_symlink, .vop_write = ncl_write, .vop_getacl = nfs_getacl, .vop_setacl = nfs_setacl, .vop_set_text = nfs_set_text, }; struct vop_vector newnfs_fifoops = { .vop_default = &fifo_specops, .vop_access = nfsspec_access, .vop_close = nfsfifo_close, .vop_fsync = nfs_fsync, .vop_getattr = nfs_getattr, .vop_inactive = ncl_inactive, .vop_print = nfs_print, .vop_read = nfsfifo_read, .vop_reclaim = ncl_reclaim, .vop_setattr = nfs_setattr, .vop_write = nfsfifo_write, }; static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap); static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, int namelen, struct ucred *cred, struct thread *td); static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td); static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, struct sillyrename *sp); /* * Global variables */ SYSCTL_DECL(_vfs_nfs); static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO; SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW, &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout"); static int nfs_prime_access_cache = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW, &nfs_prime_access_cache, 0, "Prime NFS ACCESS cache when fetching attributes"); static int newnfs_commit_on_close = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW, &newnfs_commit_on_close, 0, "write+commit on close, else only write"); static int nfs_clean_pages_on_close = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW, &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close"); int newnfs_directio_enable = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW, &newnfs_directio_enable, 0, "Enable NFS directio"); int nfs_keep_dirty_on_error; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW, &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned"); /* * This sysctl allows other processes to mmap a file that has been opened * O_DIRECT by a process. In general, having processes mmap the file while * Direct IO is in progress can lead to Data Inconsistencies. But, we allow * this by default to prevent DoS attacks - to prevent a malicious user from * opening up files O_DIRECT preventing other users from mmap'ing these * files. "Protected" environments where stricter consistency guarantees are * required can disable this knob. The process that opened the file O_DIRECT * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not * meaningful. */ int newnfs_directio_allow_mmap = 1; SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW, &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens"); #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \ | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \ | NFSACCESS_DELETE | NFSACCESS_LOOKUP) /* * SMP Locking Note : * The list of locks after the description of the lock is the ordering * of other locks acquired with the lock held. * np->n_mtx : Protects the fields in the nfsnode. VM Object Lock VI_MTX (acquired indirectly) * nmp->nm_mtx : Protects the fields in the nfsmount. rep->r_mtx * ncl_iod_mutex : Global lock, protects shared nfsiod state. * nfs_reqq_mtx : Global lock, protects the nfs_reqq list. nmp->nm_mtx rep->r_mtx * rep->r_mtx : Protects the fields in an nfsreq. */ static int nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td, struct ucred *cred, u_int32_t *retmode) { int error = 0, attrflag, i, lrupos; u_int32_t rmode; struct nfsnode *np = VTONFS(vp); struct nfsvattr nfsva; error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag, &rmode, NULL); if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { lrupos = 0; mtx_lock(&np->n_mtx); for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { if (np->n_accesscache[i].uid == cred->cr_uid) { np->n_accesscache[i].mode = rmode; np->n_accesscache[i].stamp = time_second; break; } if (i > 0 && np->n_accesscache[i].stamp < np->n_accesscache[lrupos].stamp) lrupos = i; } if (i == NFS_ACCESSCACHESIZE) { np->n_accesscache[lrupos].uid = cred->cr_uid; np->n_accesscache[lrupos].mode = rmode; np->n_accesscache[lrupos].stamp = time_second; } mtx_unlock(&np->n_mtx); if (retmode != NULL) *retmode = rmode; KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0); } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } #ifdef KDTRACE_HOOKS if (error != 0) KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0, error); #endif return (error); } /* * nfs access vnode op. * For nfs version 2, just return ok. File accesses may fail later. * For nfs version 3, use the access rpc to check accessibility. If file modes * are changed on the server, accesses might still fail later. */ static int nfs_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; int error = 0, i, gotahit; u_int32_t mode, wmode, rmode; int v34 = NFS_ISV34(vp); struct nfsnode *np = VTONFS(vp); /* * Disallow write attempts on filesystems mounted read-only; * unless the file is a socket, fifo, or a block or character * device resident on the filesystem. */ if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS | VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL | VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: return (EROFS); default: break; } } /* * For nfs v3 or v4, check to see if we have done this recently, and if * so return our cached result instead of making an ACCESS call. * If not, do an access rpc, otherwise you are stuck emulating * ufs_access() locally using the vattr. This may not be correct, * since the server may apply other access criteria such as * client uid-->server uid mapping that we do not know about. */ if (v34) { if (ap->a_accmode & VREAD) mode = NFSACCESS_READ; else mode = 0; if (vp->v_type != VDIR) { if (ap->a_accmode & VWRITE) mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); if (ap->a_accmode & VAPPEND) mode |= NFSACCESS_EXTEND; if (ap->a_accmode & VEXEC) mode |= NFSACCESS_EXECUTE; if (ap->a_accmode & VDELETE) mode |= NFSACCESS_DELETE; } else { if (ap->a_accmode & VWRITE) mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); if (ap->a_accmode & VAPPEND) mode |= NFSACCESS_EXTEND; if (ap->a_accmode & VEXEC) mode |= NFSACCESS_LOOKUP; if (ap->a_accmode & VDELETE) mode |= NFSACCESS_DELETE; if (ap->a_accmode & VDELETE_CHILD) mode |= NFSACCESS_MODIFY; } /* XXX safety belt, only make blanket request if caching */ if (nfsaccess_cache_timeout > 0) { wmode = NFSACCESS_READ | NFSACCESS_MODIFY | NFSACCESS_EXTEND | NFSACCESS_EXECUTE | NFSACCESS_DELETE | NFSACCESS_LOOKUP; } else { wmode = mode; } /* * Does our cached result allow us to give a definite yes to * this request? */ gotahit = 0; mtx_lock(&np->n_mtx); for (i = 0; i < NFS_ACCESSCACHESIZE; i++) { if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) { if (time_second < (np->n_accesscache[i].stamp + nfsaccess_cache_timeout) && (np->n_accesscache[i].mode & mode) == mode) { NFSINCRGLOBAL(nfsstatsv1.accesscache_hits); gotahit = 1; } break; } } mtx_unlock(&np->n_mtx); #ifdef KDTRACE_HOOKS if (gotahit != 0) KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp, ap->a_cred->cr_uid, mode); else KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp, ap->a_cred->cr_uid, mode); #endif if (gotahit == 0) { /* * Either a no, or a don't know. Go to the wire. */ NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); error = nfs34_access_otw(vp, wmode, ap->a_td, ap->a_cred, &rmode); if (!error && (rmode & mode) != mode) error = EACCES; } return (error); } else { if ((error = nfsspec_access(ap)) != 0) { return (error); } /* * Attempt to prevent a mapped root from accessing a file * which it shouldn't. We try to read a byte from the file * if the user is root and the file is not zero length. * After calling nfsspec_access, we should have the correct * file size cached. */ mtx_lock(&np->n_mtx); if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD) && VTONFS(vp)->n_size > 0) { struct iovec aiov; struct uio auio; char buf[1]; mtx_unlock(&np->n_mtx); aiov.iov_base = buf; aiov.iov_len = 1; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_resid = 1; auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_td = ap->a_td; if (vp->v_type == VREG) error = ncl_readrpc(vp, &auio, ap->a_cred); else if (vp->v_type == VDIR) { char* bp; bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK); aiov.iov_base = bp; aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ; error = ncl_readdirrpc(vp, &auio, ap->a_cred, ap->a_td); free(bp, M_TEMP); } else if (vp->v_type == VLNK) error = ncl_readlinkrpc(vp, &auio, ap->a_cred); else error = EACCES; } else mtx_unlock(&np->n_mtx); return (error); } } /* * nfs open vnode op * Check to see if the type is ok * and that deletion is not in progress. * For paged in text files, you will need to flush the page cache * if consistency is lost. */ /* ARGSUSED */ static int nfs_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct vattr vattr; int error; int fmode = ap->a_mode; struct ucred *cred; if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK) return (EOPNOTSUPP); /* * For NFSv4, we need to do the Open Op before cache validation, * so that we conform to RFC3530 Sec. 9.3.1. */ if (NFS_ISV4(vp)) { error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td); if (error) { error = nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); return (error); } } /* * Now, if this Open will be doing reading, re-validate/flush the * cache, so that Close/Open coherency is maintained. */ mtx_lock(&np->n_mtx); if (np->n_flag & NMODIFIED) { mtx_unlock(&np->n_mtx); error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) return (EBADF); if (error == EINTR || error == EIO) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } mtx_lock(&np->n_mtx); np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); if (vp->v_type == VDIR) np->n_direofoffset = 0; mtx_unlock(&np->n_mtx); error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } mtx_lock(&np->n_mtx); np->n_mtime = vattr.va_mtime; if (NFS_ISV4(vp)) np->n_change = vattr.va_filerev; } else { mtx_unlock(&np->n_mtx); error = VOP_GETATTR(vp, &vattr, ap->a_cred); if (error) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } mtx_lock(&np->n_mtx); if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) || NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { if (vp->v_type == VDIR) np->n_direofoffset = 0; mtx_unlock(&np->n_mtx); error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) return (EBADF); if (error == EINTR || error == EIO) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } mtx_lock(&np->n_mtx); np->n_mtime = vattr.va_mtime; if (NFS_ISV4(vp)) np->n_change = vattr.va_filerev; } } /* * If the object has >= 1 O_DIRECT active opens, we disable caching. */ if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { if (np->n_directio_opens == 0) { mtx_unlock(&np->n_mtx); error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) return (EBADF); if (error) { if (NFS_ISV4(vp)) (void) nfsrpc_close(vp, 0, ap->a_td); return (error); } mtx_lock(&np->n_mtx); np->n_flag |= NNONCACHE; } np->n_directio_opens++; } /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */ if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0) np->n_flag |= NWRITEOPENED; /* * If this is an open for writing, capture a reference to the * credentials, so they can be used by ncl_putpages(). Using * these write credentials is preferable to the credentials of * whatever thread happens to be doing the VOP_PUTPAGES() since * the write RPCs are less likely to fail with EACCES. */ if ((fmode & FWRITE) != 0) { cred = np->n_writecred; np->n_writecred = crhold(ap->a_cred); } else cred = NULL; mtx_unlock(&np->n_mtx); if (cred != NULL) crfree(cred); vnode_create_vobject(vp, vattr.va_size, ap->a_td); return (0); } /* * nfs close vnode op * What an NFS client should do upon close after writing is a debatable issue. * Most NFS clients push delayed writes to the server upon close, basically for * two reasons: * 1 - So that any write errors may be reported back to the client process * doing the close system call. By far the two most likely errors are * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure. * 2 - To put a worst case upper bound on cache inconsistency between * multiple clients for the file. * There is also a consistency problem for Version 2 of the protocol w.r.t. * not being able to tell if other clients are writing a file concurrently, * since there is no way of knowing if the changed modify time in the reply * is only due to the write for this client. * (NFS Version 3 provides weak cache consistency data in the reply that * should be sufficient to detect and handle this case.) * * The current code does the following: * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers * for NFS Version 3 - flush dirty buffers to the server but don't invalidate * or commit them (this satisfies 1 and 2 except for the * case where the server crashes after this close but * before the commit RPC, which is felt to be "good * enough". Changing the last argument to ncl_flush() to * a 1 would force a commit operation, if it is felt a * commit is necessary now. * for NFS Version 4 - flush the dirty buffers and commit them, if * nfscl_mustflush() says this is necessary. * It is necessary if there is no write delegation held, * in order to satisfy open/close coherency. * If the file isn't cached on local stable storage, * it may be necessary in order to detect "out of space" * errors from the server, if the write delegation * issued by the server doesn't allow the file to grow. */ /* ARGSUSED */ static int nfs_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct nfsvattr nfsva; struct ucred *cred; int error = 0, ret, localcred = 0; int fmode = ap->a_fflag; - if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF)) + if (NFSCL_FORCEDISM(vp->v_mount)) return (0); /* * During shutdown, a_cred isn't valid, so just use root. */ if (ap->a_cred == NOCRED) { cred = newnfs_getcred(); localcred = 1; } else { cred = ap->a_cred; } if (vp->v_type == VREG) { /* * Examine and clean dirty pages, regardless of NMODIFIED. * This closes a major hole in close-to-open consistency. * We want to push out all dirty pages (and buffers) on * close, regardless of whether they were dirtied by * mmap'ed writes or via write(). */ if (nfs_clean_pages_on_close && vp->v_object) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } mtx_lock(&np->n_mtx); if (np->n_flag & NMODIFIED) { mtx_unlock(&np->n_mtx); if (NFS_ISV3(vp)) { /* * Under NFSv3 we have dirty buffers to dispose of. We * must flush them to the NFS server. We have the option * of waiting all the way through the commit rpc or just * waiting for the initial write. The default is to only * wait through the initial write so the data is in the * server's cache, which is roughly similar to the state * a standard disk subsystem leaves the file in on close(). * * We cannot clear the NMODIFIED bit in np->n_flag due to * potential races with other processes, and certainly * cannot clear it if we don't commit. * These races occur when there is no longer the old * traditional vnode locking implemented for Vnode Ops. */ int cm = newnfs_commit_on_close ? 1 : 0; error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); /* np->n_flag &= ~NMODIFIED; */ } else if (NFS_ISV4(vp)) { if (nfscl_mustflush(vp) != 0) { int cm = newnfs_commit_on_close ? 1 : 0; error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0); /* * as above w.r.t races when clearing * NMODIFIED. * np->n_flag &= ~NMODIFIED; */ } } else { error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) return (EBADF); } mtx_lock(&np->n_mtx); } /* * Invalidate the attribute cache in all cases. * An open is going to fetch fresh attrs any way, other procs * on this node that have file open will be forced to do an * otw attr fetch, but this is safe. * --> A user found that their RPC count dropped by 20% when * this was commented out and I can't see any requirement * for it, so I've disabled it when negative lookups are * enabled. (What does this have to do with negative lookup * caching? Well nothing, except it was reported by the * same user that needed negative lookup caching and I wanted * there to be a way to disable it to see if it * is the cause of some caching/coherency issue that might * crop up.) */ if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) { np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } if (np->n_flag & NWRITEERR) { np->n_flag &= ~NWRITEERR; error = np->n_error; } mtx_unlock(&np->n_mtx); } if (NFS_ISV4(vp)) { /* * Get attributes so "change" is up to date. */ if (error == 0 && nfscl_mustflush(vp) != 0 && vp->v_type == VREG && (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) { ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva, NULL); if (!ret) { np->n_change = nfsva.na_filerev; (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 0); } } /* * and do the close. */ ret = nfsrpc_close(vp, 0, ap->a_td); if (!error && ret) error = ret; if (error) error = nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); } if (newnfs_directio_enable) KASSERT((np->n_directio_asyncwr == 0), ("nfs_close: dirty unflushed (%d) directio buffers\n", np->n_directio_asyncwr)); if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) { mtx_lock(&np->n_mtx); KASSERT((np->n_directio_opens > 0), ("nfs_close: unexpectedly value (0) of n_directio_opens\n")); np->n_directio_opens--; if (np->n_directio_opens == 0) np->n_flag &= ~NNONCACHE; mtx_unlock(&np->n_mtx); } if (localcred) NFSFREECRED(cred); return (error); } /* * nfs getattr call from vfs. */ static int nfs_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = curthread; /* XXX */ struct nfsnode *np = VTONFS(vp); int error = 0; struct nfsvattr nfsva; struct vattr *vap = ap->a_vap; struct vattr vattr; /* * Update local times for special files. */ mtx_lock(&np->n_mtx); if (np->n_flag & (NACC | NUPD)) np->n_flag |= NCHG; mtx_unlock(&np->n_mtx); /* * First look in the cache. */ if (ncl_getattrcache(vp, &vattr) == 0) { vap->va_type = vattr.va_type; vap->va_mode = vattr.va_mode; vap->va_nlink = vattr.va_nlink; vap->va_uid = vattr.va_uid; vap->va_gid = vattr.va_gid; vap->va_fsid = vattr.va_fsid; vap->va_fileid = vattr.va_fileid; vap->va_size = vattr.va_size; vap->va_blocksize = vattr.va_blocksize; vap->va_atime = vattr.va_atime; vap->va_mtime = vattr.va_mtime; vap->va_ctime = vattr.va_ctime; vap->va_gen = vattr.va_gen; vap->va_flags = vattr.va_flags; vap->va_rdev = vattr.va_rdev; vap->va_bytes = vattr.va_bytes; vap->va_filerev = vattr.va_filerev; /* * Get the local modify time for the case of a write * delegation. */ nfscl_deleggetmodtime(vp, &vap->va_mtime); return (0); } if (NFS_ISV34(vp) && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) { NFSINCRGLOBAL(nfsstatsv1.accesscache_misses); nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL); if (ncl_getattrcache(vp, ap->a_vap) == 0) { nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime); return (0); } } error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL); if (!error) error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0); if (!error) { /* * Get the local modify time for the case of a write * delegation. */ nfscl_deleggetmodtime(vp, &vap->va_mtime); } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } return (error); } /* * nfs setattr call. */ static int nfs_setattr(struct vop_setattr_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct thread *td = curthread; /* XXX */ struct vattr *vap = ap->a_vap; int error = 0; u_quad_t tsize; #ifndef nolint tsize = (u_quad_t)0; #endif /* * Setting of flags and marking of atimes are not supported. */ if (vap->va_flags != VNOVAL) return (EOPNOTSUPP); /* * Disallow write attempts if the filesystem is mounted read-only. */ if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) && (vp->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); if (vap->va_size != VNOVAL) { switch (vp->v_type) { case VDIR: return (EISDIR); case VCHR: case VBLK: case VSOCK: case VFIFO: if (vap->va_mtime.tv_sec == VNOVAL && vap->va_atime.tv_sec == VNOVAL && vap->va_mode == (mode_t)VNOVAL && vap->va_uid == (uid_t)VNOVAL && vap->va_gid == (gid_t)VNOVAL) return (0); vap->va_size = VNOVAL; break; default: /* * Disallow write attempts if the filesystem is * mounted read-only. */ if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * We run vnode_pager_setsize() early (why?), * we must set np->n_size now to avoid vinvalbuf * V_SAVE races that might setsize a lower * value. */ mtx_lock(&np->n_mtx); tsize = np->n_size; mtx_unlock(&np->n_mtx); error = ncl_meta_setsize(vp, ap->a_cred, td, vap->va_size); mtx_lock(&np->n_mtx); if (np->n_flag & NMODIFIED) { tsize = np->n_size; mtx_unlock(&np->n_mtx); error = ncl_vinvalbuf(vp, vap->va_size == 0 ? 0 : V_SAVE, td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) error = EBADF; if (error != 0) { vnode_pager_setsize(vp, tsize); return (error); } /* * Call nfscl_delegmodtime() to set the modify time * locally, as required. */ nfscl_delegmodtime(vp); } else mtx_unlock(&np->n_mtx); /* * np->n_size has already been set to vap->va_size * in ncl_meta_setsize(). We must set it again since * nfs_loadattrcache() could be called through * ncl_meta_setsize() and could modify np->n_size. */ mtx_lock(&np->n_mtx); np->n_vattr.na_size = np->n_size = vap->va_size; mtx_unlock(&np->n_mtx); } } else { mtx_lock(&np->n_mtx); if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) && vp->v_type == VREG) { mtx_unlock(&np->n_mtx); error = ncl_vinvalbuf(vp, V_SAVE, td, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) return (EBADF); if (error == EINTR || error == EIO) return (error); } else mtx_unlock(&np->n_mtx); } error = nfs_setattrrpc(vp, vap, ap->a_cred, td); if (error && vap->va_size != VNOVAL) { mtx_lock(&np->n_mtx); np->n_size = np->n_vattr.na_size = tsize; vnode_pager_setsize(vp, tsize); mtx_unlock(&np->n_mtx); } return (error); } /* * Do an nfs setattr rpc. */ static int nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { struct nfsnode *np = VTONFS(vp); int error, ret, attrflag, i; struct nfsvattr nfsva; if (NFS_ISV34(vp)) { mtx_lock(&np->n_mtx); for (i = 0; i < NFS_ACCESSCACHESIZE; i++) np->n_accesscache[i].stamp = 0; np->n_flag |= NDELEGMOD; mtx_unlock(&np->n_mtx); KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp); } error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag, NULL); if (attrflag) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (error && NFS_ISV4(vp)) error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid); return (error); } /* * nfs lookup call, one step at a time... * First look in cache * If not found, unlock the directory nfsnode and do the rpc */ static int nfs_lookup(struct vop_lookup_args *ap) { struct componentname *cnp = ap->a_cnp; struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct mount *mp = dvp->v_mount; int flags = cnp->cn_flags; struct vnode *newvp; struct nfsmount *nmp; struct nfsnode *np, *newnp; int error = 0, attrflag, dattrflag, ltype, ncticks; struct thread *td = cnp->cn_thread; struct nfsfh *nfhp; struct nfsvattr dnfsva, nfsva; struct vattr vattr; struct timespec nctime; *vpp = NULLVP; if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); if (dvp->v_type != VDIR) return (ENOTDIR); nmp = VFSTONFS(mp); np = VTONFS(dvp); /* For NFSv4, wait until any remove is done. */ mtx_lock(&np->n_mtx); while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) { np->n_flag |= NREMOVEWANT; (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0); } mtx_unlock(&np->n_mtx); if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) return (error); error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks); if (error > 0 && error != ENOENT) return (error); if (error == -1) { /* * Lookups of "." are special and always return the * current directory. cache_lookup() already handles * associated locking bookkeeping, etc. */ if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { /* XXX: Is this really correct? */ if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) cnp->cn_flags |= SAVENAME; return (0); } /* * We only accept a positive hit in the cache if the * change time of the file matches our cached copy. * Otherwise, we discard the cache entry and fallback * to doing a lookup RPC. We also only trust cache * entries for less than nm_nametimeo seconds. * * To better handle stale file handles and attributes, * clear the attribute cache of this node if it is a * leaf component, part of an open() call, and not * locally modified before fetching the attributes. * This should allow stale file handles to be detected * here where we can fall back to a LOOKUP RPC to * recover rather than having nfs_open() detect the * stale file handle and failing open(2) with ESTALE. */ newvp = *vpp; newnp = VTONFS(newvp); if (!(nmp->nm_flag & NFSMNT_NOCTO) && (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && !(newnp->n_flag & NMODIFIED)) { mtx_lock(&newnp->n_mtx); newnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); mtx_unlock(&newnp->n_mtx); } if (nfscl_nodeleg(newvp, 0) == 0 || ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) && VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 && timespeccmp(&vattr.va_ctime, &nctime, ==))) { NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) cnp->cn_flags |= SAVENAME; return (0); } cache_purge(newvp); if (dvp != newvp) vput(newvp); else vrele(newvp); *vpp = NULLVP; } else if (error == ENOENT) { if (dvp->v_iflag & VI_DOOMED) return (ENOENT); /* * We only accept a negative hit in the cache if the * modification time of the parent directory matches * the cached copy in the name cache entry. * Otherwise, we discard all of the negative cache * entries for this directory. We also only trust * negative cache entries for up to nm_negnametimeo * seconds. */ if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) && VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 && timespeccmp(&vattr.va_mtime, &nctime, ==)) { NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits); return (ENOENT); } cache_purge_negative(dvp); } error = 0; newvp = NULLVP; NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses); error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (error) { if (newvp != NULLVP) { vput(newvp); *vpp = NULLVP; } if (error != ENOENT) { if (NFS_ISV4(dvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* The requested file was not found. */ if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) && (flags & ISLASTCN)) { /* * XXX: UFS does a full VOP_ACCESS(dvp, * VWRITE) here instead of just checking * MNT_RDONLY. */ if (mp->mnt_flag & MNT_RDONLY) return (EROFS); cnp->cn_flags |= SAVENAME; return (EJUSTRETURN); } if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) { /* * Cache the modification time of the parent * directory from the post-op attributes in * the name cache entry. The negative cache * entry will be ignored once the directory * has changed. Don't bother adding the entry * if the directory has already changed. */ mtx_lock(&np->n_mtx); if (timespeccmp(&np->n_vattr.na_mtime, &dnfsva.na_mtime, ==)) { mtx_unlock(&np->n_mtx); cache_enter_time(dvp, NULL, cnp, &dnfsva.na_mtime, NULL); } else mtx_unlock(&np->n_mtx); } return (ENOENT); } /* * Handle RENAME case... */ if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) { if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { FREE((caddr_t)nfhp, M_NFSFH); return (EISDIR); } error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, LK_EXCLUSIVE); if (error) return (error); newvp = NFSTOV(np); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); *vpp = newvp; cnp->cn_flags |= SAVENAME; return (0); } if (flags & ISDOTDOT) { ltype = NFSVOPISLOCKED(dvp); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); NFSVOPUNLOCK(dvp, 0); error = vfs_busy(mp, 0); NFSVOPLOCK(dvp, ltype | LK_RETRY); vfs_rel(mp); if (error == 0 && (dvp->v_iflag & VI_DOOMED)) { vfs_unbusy(mp); error = ENOENT; } if (error != 0) return (error); } NFSVOPUNLOCK(dvp, 0); error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, cnp->cn_lkflags); if (error == 0) newvp = NFSTOV(np); vfs_unbusy(mp); if (newvp != dvp) NFSVOPLOCK(dvp, ltype | LK_RETRY); if (dvp->v_iflag & VI_DOOMED) { if (error == 0) { if (newvp == dvp) vrele(newvp); else vput(newvp); } error = ENOENT; } if (error != 0) return (error); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) { FREE((caddr_t)nfhp, M_NFSFH); VREF(dvp); newvp = dvp; if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else { error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL, cnp->cn_lkflags); if (error) return (error); newvp = NFSTOV(np); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) && !(np->n_flag & NMODIFIED)) { /* * Flush the attribute cache when opening a * leaf node to ensure that fresh attributes * are fetched in nfs_open() since we did not * fetch attributes from the LOOKUP reply. */ mtx_lock(&np->n_mtx); np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp); mtx_unlock(&np->n_mtx); } } if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN)) cnp->cn_flags |= SAVENAME; if ((cnp->cn_flags & MAKEENTRY) && (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) && attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0)) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime); *vpp = newvp; return (0); } /* * nfs read call. * Just call ncl_bioread() to do the work. */ static int nfs_read(struct vop_read_args *ap) { struct vnode *vp = ap->a_vp; switch (vp->v_type) { case VREG: return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred)); case VDIR: return (EISDIR); default: return (EOPNOTSUPP); } } /* * nfs readlink call */ static int nfs_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; if (vp->v_type != VLNK) return (EINVAL); return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred)); } /* * Do a readlink rpc. * Called by ncl_doio() from below the buffer cache. */ int ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) { int error, ret, attrflag; struct nfsvattr nfsva; error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, NULL); if (attrflag) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (error && NFS_ISV4(vp)) error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs read rpc call * Ditto above */ int ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred) { int error, ret, attrflag; struct nfsvattr nfsva; struct nfsmount *nmp; nmp = VFSTONFS(vnode_mount(vp)); error = EIO; attrflag = 0; if (NFSHASPNFS(nmp)) error = nfscl_doiods(vp, uiop, NULL, NULL, NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td); NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error); if (error != 0) error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag, NULL); if (attrflag) { ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (error && NFS_ISV4(vp)) error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs write call */ int ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, int *iomode, int *must_commit, int called_from_strategy) { struct nfsvattr nfsva; int error, attrflag, ret; struct nfsmount *nmp; nmp = VFSTONFS(vnode_mount(vp)); error = EIO; attrflag = 0; if (NFSHASPNFS(nmp)) error = nfscl_doiods(vp, uiop, iomode, must_commit, NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td); NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error); if (error != 0) error = nfsrpc_write(vp, uiop, iomode, must_commit, cred, uiop->uio_td, &nfsva, &attrflag, NULL, called_from_strategy); if (attrflag) { if (VTONFS(vp)->n_flag & ND_NFSV4) ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1, 1); else ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (ret && !error) error = ret; } if (DOINGASYNC(vp)) *iomode = NFSWRITE_FILESYNC; if (error && NFS_ISV4(vp)) error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs mknod rpc * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the * mode set to specify the file type and the size field for rdev. */ static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct vattr *vap) { struct nfsvattr nfsva, dnfsva; struct vnode *newvp = NULL; struct nfsnode *np = NULL, *dnp; struct nfsfh *nfhp; struct vattr vattr; int error = 0, attrflag, dattrflag; u_int32_t rdev; if (vap->va_type == VCHR || vap->va_type == VBLK) rdev = vap->va_rdev; else if (vap->va_type == VFIFO || vap->va_type == VSOCK) rdev = 0xffffffff; else return (EOPNOTSUPP); if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) return (error); error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (!error) { if (!nfhp) (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (nfhp) error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); } if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (!error) { newvp = NFSTOV(np); if (attrflag != 0) { error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); if (error != 0) vput(newvp); } } if (!error) { *vpp = newvp; } else if (NFS_ISV4(dvp)) { error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, vap->va_gid); } dnp = VTONFS(dvp); mtx_lock(&dnp->n_mtx); dnp->n_flag |= NMODIFIED; if (!dattrflag) { dnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } mtx_unlock(&dnp->n_mtx); return (error); } /* * nfs mknod vop * just call nfs_mknodrpc() to do the work. */ /* ARGSUSED */ static int nfs_mknod(struct vop_mknod_args *ap) { return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap)); } static struct mtx nfs_cverf_mtx; MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex", MTX_DEF); static nfsquad_t nfs_get_cverf(void) { static nfsquad_t cverf; nfsquad_t ret; static int cverf_initialized = 0; mtx_lock(&nfs_cverf_mtx); if (cverf_initialized == 0) { cverf.lval[0] = arc4random(); cverf.lval[1] = arc4random(); cverf_initialized = 1; } else cverf.qval++; ret = cverf; mtx_unlock(&nfs_cverf_mtx); return (ret); } /* * nfs file create call */ static int nfs_create(struct vop_create_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct nfsnode *np = NULL, *dnp; struct vnode *newvp = NULL; struct nfsmount *nmp; struct nfsvattr dnfsva, nfsva; struct nfsfh *nfhp; nfsquad_t cverf; int error = 0, attrflag, dattrflag, fmode = 0; struct vattr vattr; /* * Oops, not for me.. */ if (vap->va_type == VSOCK) return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap)); if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred))) return (error); if (vap->va_vaflags & VA_EXCLUSIVE) fmode |= O_EXCL; dnp = VTONFS(dvp); nmp = VFSTONFS(vnode_mount(dvp)); again: /* For NFSv4, wait until any remove is done. */ mtx_lock(&dnp->n_mtx); while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) { dnp->n_flag |= NREMOVEWANT; (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0); } mtx_unlock(&dnp->n_mtx); cverf = nfs_get_cverf(); error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (!error) { if (nfhp == NULL) (void) nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (nfhp != NULL) error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); } if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (!error) { newvp = NFSTOV(np); if (attrflag == 0) error = nfsrpc_getattr(newvp, cnp->cn_cred, cnp->cn_thread, &nfsva, NULL); if (error == 0) error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } if (error) { if (newvp != NULL) { vput(newvp); newvp = NULL; } if (NFS_ISV34(dvp) && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) { fmode &= ~O_EXCL; goto again; } } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) { if (nfscl_checksattr(vap, &nfsva)) { error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, cnp->cn_thread, &nfsva, &attrflag, NULL); if (error && (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL)) { /* try again without setting uid/gid */ vap->va_uid = (uid_t)VNOVAL; vap->va_gid = (uid_t)VNOVAL; error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred, cnp->cn_thread, &nfsva, &attrflag, NULL); } if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); if (error != 0) vput(newvp); } } if (!error) { if ((cnp->cn_flags & MAKEENTRY) && attrflag) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); *ap->a_vpp = newvp; } else if (NFS_ISV4(dvp)) { error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, vap->va_gid); } mtx_lock(&dnp->n_mtx); dnp->n_flag |= NMODIFIED; if (!dattrflag) { dnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } mtx_unlock(&dnp->n_mtx); return (error); } /* * nfs file remove call * To try and make nfs semantics closer to ufs semantics, a file that has * other processes using the vnode is renamed instead of removed and then * removed later on the last close. * - If v_usecount > 1 * If a rename is not already in the works * call nfs_sillyrename() to set it up * else * do the remove rpc */ static int nfs_remove(struct vop_remove_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct nfsnode *np = VTONFS(vp); int error = 0; struct vattr vattr; KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name")); KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount")); if (vp->v_type == VDIR) error = EPERM; else if (vrefcnt(vp) == 1 || (np->n_sillyrename && VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 && vattr.va_nlink > 1)) { /* * Purge the name cache so that the chance of a lookup for * the name succeeding while the remove is in progress is * minimized. Without node locking it can still happen, such * that an I/O op returns ESTALE, but since you get this if * another host removes the file.. */ cache_purge(vp); /* * throw away biocache buffers, mainly to avoid * unnecessary delayed writes later. */ error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1); if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) error = EBADF; else if (error != EINTR && error != EIO) /* Do the rpc */ error = nfs_removerpc(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread); /* * Kludge City: If the first reply to the remove rpc is lost.. * the reply to the retransmitted request will be ENOENT * since the file was in fact removed * Therefore, we cheat and return success. */ if (error == ENOENT) error = 0; } else if (!np->n_sillyrename) error = nfs_sillyrename(dvp, vp, cnp); mtx_lock(&np->n_mtx); np->n_attrstamp = 0; mtx_unlock(&np->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); return (error); } /* * nfs file remove rpc called from nfs_inactive */ int ncl_removeit(struct sillyrename *sp, struct vnode *vp) { /* * Make sure that the directory vnode is still valid. * XXX we should lock sp->s_dvp here. */ if (sp->s_dvp->v_type == VBAD) return (0); return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen, sp->s_cred, NULL)); } /* * Nfs remove rpc, called from nfs_remove() and ncl_removeit(). */ static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name, int namelen, struct ucred *cred, struct thread *td) { struct nfsvattr dnfsva; struct nfsnode *dnp = VTONFS(dvp); int error = 0, dattrflag; mtx_lock(&dnp->n_mtx); dnp->n_flag |= NREMOVEINPROG; mtx_unlock(&dnp->n_mtx); error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva, &dattrflag, NULL); mtx_lock(&dnp->n_mtx); if ((dnp->n_flag & NREMOVEWANT)) { dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG); mtx_unlock(&dnp->n_mtx); wakeup((caddr_t)dnp); } else { dnp->n_flag &= ~NREMOVEINPROG; mtx_unlock(&dnp->n_mtx); } if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); mtx_lock(&dnp->n_mtx); dnp->n_flag |= NMODIFIED; if (!dattrflag) { dnp->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } mtx_unlock(&dnp->n_mtx); if (error && NFS_ISV4(dvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs file rename call */ static int nfs_rename(struct vop_rename_args *ap) { struct vnode *fvp = ap->a_fvp; struct vnode *tvp = ap->a_tvp; struct vnode *fdvp = ap->a_fdvp; struct vnode *tdvp = ap->a_tdvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct nfsnode *fnp = VTONFS(ap->a_fvp); struct nfsnode *tdnp = VTONFS(ap->a_tdvp); struct nfsv4node *newv4 = NULL; int error; KASSERT((tcnp->cn_flags & HASBUF) != 0 && (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name")); /* Check for cross-device rename */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; goto out; } if (fvp == tvp) { printf("nfs_rename: fvp == tvp (can't happen)\n"); error = 0; goto out; } if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0) goto out; /* * We have to flush B_DELWRI data prior to renaming * the file. If we don't, the delayed-write buffers * can be flushed out later after the file has gone stale * under NFSV3. NFSV2 does not have this problem because * ( as far as I can tell ) it flushes dirty buffers more * often. * * Skip the rename operation if the fsync fails, this can happen * due to the server's volume being full, when we pushed out data * that was written back to our cache earlier. Not checking for * this condition can result in potential (silent) data loss. */ error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread); NFSVOPUNLOCK(fvp, 0); if (!error && tvp) error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread); if (error) goto out; /* * If the tvp exists and is in use, sillyrename it before doing the * rename of the new file over it. * XXX Can't sillyrename a directory. */ if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename && tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) { vput(tvp); tvp = NULL; } error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen, tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred, tcnp->cn_thread); if (error == 0 && NFS_ISV4(tdvp)) { /* * For NFSv4, check to see if it is the same name and * replace the name, if it is different. */ MALLOC(newv4, struct nfsv4node *, sizeof (struct nfsv4node) + tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK); mtx_lock(&tdnp->n_mtx); mtx_lock(&fnp->n_mtx); if (fnp->n_v4 != NULL && fvp->v_type == VREG && (fnp->n_v4->n4_namelen != tcnp->cn_namelen || NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen) || tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen || NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, tdnp->n_fhp->nfh_len))) { #ifdef notdef { char nnn[100]; int nnnl; nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99; bcopy(tcnp->cn_nameptr, nnn, nnnl); nnn[nnnl] = '\0'; printf("ren replace=%s\n",nnn); } #endif FREE((caddr_t)fnp->n_v4, M_NFSV4NODE); fnp->n_v4 = newv4; newv4 = NULL; fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len; fnp->n_v4->n4_namelen = tcnp->cn_namelen; NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data, tdnp->n_fhp->nfh_len); NFSBCOPY(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen); } mtx_unlock(&tdnp->n_mtx); mtx_unlock(&fnp->n_mtx); if (newv4 != NULL) FREE((caddr_t)newv4, M_NFSV4NODE); } if (fvp->v_type == VDIR) { if (tvp != NULL && tvp->v_type == VDIR) cache_purge(tdvp); cache_purge(fdvp); } out: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); /* * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry. */ if (error == ENOENT) error = 0; return (error); } /* * nfs file rename rpc called from nfs_remove() above */ static int nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp, struct sillyrename *sp) { return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen, sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread)); } /* * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit(). */ static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td) { struct nfsvattr fnfsva, tnfsva; struct nfsnode *fdnp = VTONFS(fdvp); struct nfsnode *tdnp = VTONFS(tdvp); int error = 0, fattrflag, tattrflag; error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp, tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag, &tattrflag, NULL, NULL); mtx_lock(&fdnp->n_mtx); fdnp->n_flag |= NMODIFIED; if (fattrflag != 0) { mtx_unlock(&fdnp->n_mtx); (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1); } else { fdnp->n_attrstamp = 0; mtx_unlock(&fdnp->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp); } mtx_lock(&tdnp->n_mtx); tdnp->n_flag |= NMODIFIED; if (tattrflag != 0) { mtx_unlock(&tdnp->n_mtx); (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1); } else { tdnp->n_attrstamp = 0; mtx_unlock(&tdnp->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); } if (error && NFS_ISV4(fdvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs hard link create call */ static int nfs_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct nfsnode *np, *tdnp; struct nfsvattr nfsva, dnfsva; int error = 0, attrflag, dattrflag; /* * Push all writes to the server, so that the attribute cache * doesn't get "out of sync" with the server. * XXX There should be a better way! */ VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread); error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag, &dattrflag, NULL); tdnp = VTONFS(tdvp); mtx_lock(&tdnp->n_mtx); tdnp->n_flag |= NMODIFIED; if (dattrflag != 0) { mtx_unlock(&tdnp->n_mtx); (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1); } else { tdnp->n_attrstamp = 0; mtx_unlock(&tdnp->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp); } if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); else { np = VTONFS(vp); mtx_lock(&np->n_mtx); np->n_attrstamp = 0; mtx_unlock(&np->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); } /* * If negative lookup caching is enabled, I might as well * add an entry for this node. Not necessary for correctness, * but if negative caching is enabled, then the system * must care about lookup caching hit rate, so... */ if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 && (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL); } if (error && NFS_ISV4(vp)) error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, (gid_t)0); return (error); } /* * nfs symbolic link create call */ static int nfs_symlink(struct vop_symlink_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct nfsvattr nfsva, dnfsva; struct nfsfh *nfhp; struct nfsnode *np = NULL, *dnp; struct vnode *newvp = NULL; int error = 0, attrflag, dattrflag, ret; vap->va_type = VLNK; error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen, ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (nfhp) { ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); if (!ret) newvp = NFSTOV(np); else if (!error) error = ret; } if (newvp != NULL) { if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else if (!error) { /* * If we do not have an error and we could not extract the * newvp from the response due to the request being NFSv2, we * have to do a lookup in order to obtain a newvp to return. */ error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); if (!error) newvp = NFSTOV(np); } if (error) { if (newvp) vput(newvp); if (NFS_ISV4(dvp)) error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, vap->va_gid); } else { *ap->a_vpp = newvp; } dnp = VTONFS(dvp); mtx_lock(&dnp->n_mtx); dnp->n_flag |= NMODIFIED; if (dattrflag != 0) { mtx_unlock(&dnp->n_mtx); (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); } else { dnp->n_attrstamp = 0; mtx_unlock(&dnp->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } /* * If negative lookup caching is enabled, I might as well * add an entry for this node. Not necessary for correctness, * but if negative caching is enabled, then the system * must care about lookup caching hit rate, so... */ if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) { cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL); } return (error); } /* * nfs make dir call */ static int nfs_mkdir(struct vop_mkdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct nfsnode *np = NULL, *dnp; struct vnode *newvp = NULL; struct vattr vattr; struct nfsfh *nfhp; struct nfsvattr nfsva, dnfsva; int error = 0, attrflag, dattrflag, ret; if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) return (error); vap->va_type = VDIR; error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); dnp = VTONFS(dvp); mtx_lock(&dnp->n_mtx); dnp->n_flag |= NMODIFIED; if (dattrflag != 0) { mtx_unlock(&dnp->n_mtx); (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); } else { dnp->n_attrstamp = 0; mtx_unlock(&dnp->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } if (nfhp) { ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread, &np, NULL, LK_EXCLUSIVE); if (!ret) { newvp = NFSTOV(np); if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } else if (!error) error = ret; } if (!error && newvp == NULL) { error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np); if (!error) { newvp = NFSTOV(np); if (newvp->v_type != VDIR) error = EEXIST; } } if (error) { if (newvp) vput(newvp); if (NFS_ISV4(dvp)) error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid, vap->va_gid); } else { /* * If negative lookup caching is enabled, I might as well * add an entry for this node. Not necessary for correctness, * but if negative caching is enabled, then the system * must care about lookup caching hit rate, so... */ if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 && (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && dattrflag != 0) cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, &dnfsva.na_ctime); *ap->a_vpp = newvp; } return (error); } /* * nfs remove directory call */ static int nfs_rmdir(struct vop_rmdir_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct nfsnode *dnp; struct nfsvattr dnfsva; int error, dattrflag; if (dvp == vp) return (EINVAL); error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL); dnp = VTONFS(dvp); mtx_lock(&dnp->n_mtx); dnp->n_flag |= NMODIFIED; if (dattrflag != 0) { mtx_unlock(&dnp->n_mtx); (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); } else { dnp->n_attrstamp = 0; mtx_unlock(&dnp->n_mtx); KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp); } cache_purge(dvp); cache_purge(vp); if (error && NFS_ISV4(dvp)) error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0, (gid_t)0); /* * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. */ if (error == ENOENT) error = 0; return (error); } /* * nfs readdir call */ static int nfs_readdir(struct vop_readdir_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct uio *uio = ap->a_uio; ssize_t tresid, left; int error = 0; struct vattr vattr; if (ap->a_eofflag != NULL) *ap->a_eofflag = 0; if (vp->v_type != VDIR) return(EPERM); /* * First, check for hit on the EOF offset cache */ if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset && (np->n_flag & NMODIFIED) == 0) { if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) { mtx_lock(&np->n_mtx); if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) || !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) { mtx_unlock(&np->n_mtx); NFSINCRGLOBAL(nfsstatsv1.direofcache_hits); if (ap->a_eofflag != NULL) *ap->a_eofflag = 1; return (0); } else mtx_unlock(&np->n_mtx); } } /* * NFS always guarantees that directory entries don't straddle * DIRBLKSIZ boundaries. As such, we need to limit the size * to an exact multiple of DIRBLKSIZ, to avoid copying a partial * directory entry. */ left = uio->uio_resid % DIRBLKSIZ; if (left == uio->uio_resid) return (EINVAL); uio->uio_resid -= left; /* * Call ncl_bioread() to do the real work. */ tresid = uio->uio_resid; error = ncl_bioread(vp, uio, 0, ap->a_cred); if (!error && uio->uio_resid == tresid) { NFSINCRGLOBAL(nfsstatsv1.direofcache_misses); if (ap->a_eofflag != NULL) *ap->a_eofflag = 1; } /* Add the partial DIRBLKSIZ (left) back in. */ uio->uio_resid += left; return (error); } /* * Readdir rpc call. * Called from below the buffer cache by ncl_doio(). */ int ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, struct thread *td) { struct nfsvattr nfsva; nfsuint64 *cookiep, cookie; struct nfsnode *dnp = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, eof, attrflag; KASSERT(uiop->uio_iovcnt == 1 && (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, ("nfs readdirrpc bad uio")); /* * If there is no cookie, assume directory was stale. */ ncl_dircookie_lock(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); if (cookiep) { cookie = *cookiep; ncl_dircookie_unlock(dnp); } else { ncl_dircookie_unlock(dnp); return (NFSERR_BAD_COOKIE); } if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) (void)ncl_fsinfo(nmp, vp, cred, td); error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva, &attrflag, &eof, NULL); if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { /* * We are now either at the end of the directory or have filled * the block. */ if (eof) dnp->n_direofoffset = uiop->uio_offset; else { if (uiop->uio_resid > 0) printf("EEK! readdirrpc resid > 0\n"); ncl_dircookie_lock(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); *cookiep = cookie; ncl_dircookie_unlock(dnp); } } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } return (error); } /* * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc(). */ int ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred, struct thread *td) { struct nfsvattr nfsva; nfsuint64 *cookiep, cookie; struct nfsnode *dnp = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, attrflag, eof; KASSERT(uiop->uio_iovcnt == 1 && (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 && (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0, ("nfs readdirplusrpc bad uio")); /* * If there is no cookie, assume directory was stale. */ ncl_dircookie_lock(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0); if (cookiep) { cookie = *cookiep; ncl_dircookie_unlock(dnp); } else { ncl_dircookie_unlock(dnp); return (NFSERR_BAD_COOKIE); } if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) (void)ncl_fsinfo(nmp, vp, cred, td); error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva, &attrflag, &eof, NULL); if (attrflag) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (!error) { /* * We are now either at end of the directory or have filled the * the block. */ if (eof) dnp->n_direofoffset = uiop->uio_offset; else { if (uiop->uio_resid > 0) printf("EEK! readdirplusrpc resid > 0\n"); ncl_dircookie_lock(dnp); cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1); *cookiep = cookie; ncl_dircookie_unlock(dnp); } } else if (NFS_ISV4(vp)) { error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); } return (error); } /* * Silly rename. To make the NFS filesystem that is stateless look a little * more like the "ufs" a remove of an active vnode is translated to a rename * to a funny looking filename that is removed by nfs_inactive on the * nfsnode. There is the potential for another process on a different client * to create the same funny name between the nfs_lookitup() fails and the * nfs_rename() completes, but... */ static int nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { struct sillyrename *sp; struct nfsnode *np; int error; short pid; unsigned int lticks; cache_purge(dvp); np = VTONFS(vp); KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir")); MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename), M_NEWNFSREQ, M_WAITOK); sp->s_cred = crhold(cnp->cn_cred); sp->s_dvp = dvp; VREF(dvp); /* * Fudge together a funny name. * Changing the format of the funny name to accommodate more * sillynames per directory. * The name is now changed to .nfs...4, where ticks is * CPU ticks since boot. */ pid = cnp->cn_thread->td_proc->p_pid; lticks = (unsigned int)ticks; for ( ; ; ) { sp->s_namlen = sprintf(sp->s_name, ".nfs.%08x.%04x4.4", lticks, pid); if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, cnp->cn_thread, NULL)) break; lticks++; } error = nfs_renameit(dvp, vp, cnp, sp); if (error) goto bad; error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred, cnp->cn_thread, &np); np->n_sillyrename = sp; return (0); bad: vrele(sp->s_dvp); crfree(sp->s_cred); free((caddr_t)sp, M_NEWNFSREQ); return (error); } /* * Look up a file name and optionally either update the file handle or * allocate an nfsnode, depending on the value of npp. * npp == NULL --> just do the lookup * *npp == NULL --> allocate a new nfsnode and make sure attributes are * handled too * *npp != NULL --> update the file handle in the vnode */ static int nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred, struct thread *td, struct nfsnode **npp) { struct vnode *newvp = NULL, *vp; struct nfsnode *np, *dnp = VTONFS(dvp); struct nfsfh *nfhp, *onfhp; struct nfsvattr nfsva, dnfsva; struct componentname cn; int error = 0, attrflag, dattrflag; u_int hash; error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, NULL); if (dattrflag) (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1); if (npp && !error) { if (*npp != NULL) { np = *npp; vp = NFSTOV(np); /* * For NFSv4, check to see if it is the same name and * replace the name, if it is different. */ if (np->n_v4 != NULL && nfsva.na_type == VREG && (np->n_v4->n4_namelen != len || NFSBCMP(name, NFS4NODENAME(np->n_v4), len) || dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen || NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len))) { #ifdef notdef { char nnn[100]; int nnnl; nnnl = (len < 100) ? len : 99; bcopy(name, nnn, nnnl); nnn[nnnl] = '\0'; printf("replace=%s\n",nnn); } #endif FREE((caddr_t)np->n_v4, M_NFSV4NODE); MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len + len - 1, M_NFSV4NODE, M_WAITOK); np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len; np->n_v4->n4_namelen = len; NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data, dnp->n_fhp->nfh_len); NFSBCOPY(name, NFS4NODENAME(np->n_v4), len); } hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT); onfhp = np->n_fhp; /* * Rehash node for new file handle. */ vfs_hash_rehash(vp, hash); np->n_fhp = nfhp; if (onfhp != NULL) FREE((caddr_t)onfhp, M_NFSFH); newvp = NFSTOV(np); } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) { FREE((caddr_t)nfhp, M_NFSFH); VREF(dvp); newvp = dvp; } else { cn.cn_nameptr = name; cn.cn_namelen = len; error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td, &np, NULL, LK_EXCLUSIVE); if (error) return (error); newvp = NFSTOV(np); } if (!attrflag && *npp == NULL) { if (newvp == dvp) vrele(newvp); else vput(newvp); return (ENOENT); } if (attrflag) (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 1); } if (npp && *npp == NULL) { if (error) { if (newvp) { if (newvp == dvp) vrele(newvp); else vput(newvp); } } else *npp = np; } if (error && NFS_ISV4(dvp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * Nfs Version 3 and 4 commit rpc */ int ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred, struct thread *td) { struct nfsvattr nfsva; struct nfsmount *nmp = VFSTONFS(vp->v_mount); struct nfsnode *np; struct uio uio; int error, attrflag; np = VTONFS(vp); error = EIO; attrflag = 0; if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) { uio.uio_offset = offset; uio.uio_resid = cnt; error = nfscl_doiods(vp, &uio, NULL, NULL, NFSV4OPEN_ACCESSWRITE, 1, cred, td); if (error != 0) { mtx_lock(&np->n_mtx); np->n_flag &= ~NDSCOMMIT; mtx_unlock(&np->n_mtx); } } if (error != 0) { mtx_lock(&nmp->nm_mtx); if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) { mtx_unlock(&nmp->nm_mtx); return (0); } mtx_unlock(&nmp->nm_mtx); error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva, &attrflag, NULL); } if (attrflag != 0) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error != 0 && NFS_ISV4(vp)) error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); return (error); } /* * Strategy routine. * For async requests when nfsiod(s) are running, queue the request by * calling ncl_asyncio(), otherwise just all ncl_doio() to do the * request. */ static int nfs_strategy(struct vop_strategy_args *ap) { struct buf *bp; struct vnode *vp; struct ucred *cr; bp = ap->a_bp; vp = ap->a_vp; KASSERT(bp->b_vp == vp, ("missing b_getvp")); KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp)); BUF_ASSERT_HELD(bp); if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno) bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize / DEV_BSIZE); if (bp->b_iocmd == BIO_READ) cr = bp->b_rcred; else cr = bp->b_wcred; /* * If the op is asynchronous and an i/o daemon is waiting * queue the request, wake it up and wait for completion * otherwise just do it ourselves. */ if ((bp->b_flags & B_ASYNC) == 0 || ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread)) (void) ncl_doio(vp, bp, cr, curthread, 1); return (0); } /* * fsync vnode op. Just call ncl_flush() with commit == 1. */ /* ARGSUSED */ static int nfs_fsync(struct vop_fsync_args *ap) { if (ap->a_vp->v_type != VREG) { /* * For NFS, metadata is changed synchronously on the server, * so there is nothing to flush. Also, ncl_flush() clears * the NMODIFIED flag and that shouldn't be done here for * directories. */ return (0); } return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0)); } /* * Flush all the blocks associated with a vnode. * Walk through the buffer pool and push any dirty pages * associated with the vnode. * If the called_from_renewthread argument is TRUE, it has been called * from the NFSv4 renew thread and, as such, cannot block indefinitely * waiting for a buffer write to complete. */ int ncl_flush(struct vnode *vp, int waitfor, struct thread *td, int commit, int called_from_renewthread) { struct nfsnode *np = VTONFS(vp); struct buf *bp; int i; struct buf *nbp; struct nfsmount *nmp = VFSTONFS(vp->v_mount); int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos; int passone = 1, trycnt = 0; u_quad_t off, endoff, toff; struct ucred* wcred = NULL; struct buf **bvec = NULL; struct bufobj *bo; #ifndef NFS_COMMITBVECSIZ #define NFS_COMMITBVECSIZ 20 #endif struct buf *bvec_on_stack[NFS_COMMITBVECSIZ]; int bvecsize = 0, bveccount; if (called_from_renewthread != 0) slptimeo = hz; if (nmp->nm_flag & NFSMNT_INT) slpflag = PCATCH; if (!commit) passone = 0; bo = &vp->v_bufobj; /* * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the * server, but has not been committed to stable storage on the server * yet. On the first pass, the byte range is worked out and the commit * rpc is done. On the second pass, ncl_writebp() is called to do the * job. */ again: off = (u_quad_t)-1; endoff = 0; bvecpos = 0; if (NFS_ISV34(vp) && commit) { if (bvec != NULL && bvec != bvec_on_stack) free(bvec, M_TEMP); /* * Count up how many buffers waiting for a commit. */ bveccount = 0; BO_LOCK(bo); TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (!BUF_ISLOCKED(bp) && (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) == (B_DELWRI | B_NEEDCOMMIT)) bveccount++; } /* * Allocate space to remember the list of bufs to commit. It is * important to use M_NOWAIT here to avoid a race with nfs_write. * If we can't get memory (for whatever reason), we will end up * committing the buffers one-by-one in the loop below. */ if (bveccount > NFS_COMMITBVECSIZ) { /* * Release the vnode interlock to avoid a lock * order reversal. */ BO_UNLOCK(bo); bvec = (struct buf **) malloc(bveccount * sizeof(struct buf *), M_TEMP, M_NOWAIT); BO_LOCK(bo); if (bvec == NULL) { bvec = bvec_on_stack; bvecsize = NFS_COMMITBVECSIZ; } else bvecsize = bveccount; } else { bvec = bvec_on_stack; bvecsize = NFS_COMMITBVECSIZ; } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bvecpos >= bvecsize) break; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { nbp = TAILQ_NEXT(bp, b_bobufs); continue; } if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) != (B_DELWRI | B_NEEDCOMMIT)) { BUF_UNLOCK(bp); nbp = TAILQ_NEXT(bp, b_bobufs); continue; } BO_UNLOCK(bo); bremfree(bp); /* * Work out if all buffers are using the same cred * so we can deal with them all with one commit. * * NOTE: we are not clearing B_DONE here, so we have * to do it later on in this routine if we intend to * initiate I/O on the bp. * * Note: to avoid loopback deadlocks, we do not * assign b_runningbufspace. */ if (wcred == NULL) wcred = bp->b_wcred; else if (wcred != bp->b_wcred) wcred = NOCRED; vfs_busy_pages(bp, 1); BO_LOCK(bo); /* * bp is protected by being locked, but nbp is not * and vfs_busy_pages() may sleep. We have to * recalculate nbp. */ nbp = TAILQ_NEXT(bp, b_bobufs); /* * A list of these buffers is kept so that the * second loop knows which buffers have actually * been committed. This is necessary, since there * may be a race between the commit rpc and new * uncommitted writes on the file. */ bvec[bvecpos++] = bp; toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; if (toff < off) off = toff; toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); if (toff > endoff) endoff = toff; } BO_UNLOCK(bo); } if (bvecpos > 0) { /* * Commit data on the server, as required. * If all bufs are using the same wcred, then use that with * one call for all of them, otherwise commit each one * separately. */ if (wcred != NOCRED) retv = ncl_commit(vp, off, (int)(endoff - off), wcred, td); else { retv = 0; for (i = 0; i < bvecpos; i++) { off_t off, size; bp = bvec[i]; off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE + bp->b_dirtyoff; size = (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff); retv = ncl_commit(vp, off, (int)size, bp->b_wcred, td); if (retv) break; } } if (retv == NFSERR_STALEWRITEVERF) ncl_clearcommit(vp->v_mount); /* * Now, either mark the blocks I/O done or mark the * blocks dirty, depending on whether the commit * succeeded. */ for (i = 0; i < bvecpos; i++) { bp = bvec[i]; bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK); if (retv) { /* * Error, leave B_DELWRI intact */ vfs_unbusy_pages(bp); brelse(bp); } else { /* * Success, remove B_DELWRI ( bundirty() ). * * b_dirtyoff/b_dirtyend seem to be NFS * specific. We should probably move that * into bundirty(). XXX */ bufobj_wref(bo); bp->b_flags |= B_ASYNC; bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_dirtyoff = bp->b_dirtyend = 0; bufdone(bp); } } } /* * Start/do any write(s) that are required. */ loop: BO_LOCK(bo); TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) { if (waitfor != MNT_WAIT || passone) continue; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo); if (error == 0) { BUF_UNLOCK(bp); goto loop; } if (error == ENOLCK) { error = 0; goto loop; } if (called_from_renewthread != 0) { /* * Return EIO so the flush will be retried * later. */ error = EIO; goto done; } if (newnfs_sigintr(nmp, td)) { error = EINTR; goto done; } if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } goto loop; } if ((bp->b_flags & B_DELWRI) == 0) panic("nfs_fsync: not dirty"); if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) { BUF_UNLOCK(bp); continue; } BO_UNLOCK(bo); bremfree(bp); if (passone || !commit) bp->b_flags |= B_ASYNC; else bp->b_flags |= B_ASYNC; bwrite(bp); if (newnfs_sigintr(nmp, td)) { error = EINTR; goto done; } goto loop; } if (passone) { passone = 0; BO_UNLOCK(bo); goto again; } if (waitfor == MNT_WAIT) { while (bo->bo_numoutput) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); if (called_from_renewthread != 0) { /* * Return EIO so that the flush will be * retried later. */ error = EIO; goto done; } error = newnfs_sigintr(nmp, td); if (error) goto done; if (slpflag == PCATCH) { slpflag = 0; slptimeo = 2 * hz; } BO_LOCK(bo); } } if (bo->bo_dirty.bv_cnt != 0 && commit) { BO_UNLOCK(bo); goto loop; } /* * Wait for all the async IO requests to drain */ BO_UNLOCK(bo); mtx_lock(&np->n_mtx); while (np->n_directio_asyncwr > 0) { np->n_flag |= NFSYNCWAIT; error = newnfs_msleep(td, &np->n_directio_asyncwr, &np->n_mtx, slpflag | (PRIBIO + 1), "nfsfsync", 0); if (error) { if (newnfs_sigintr(nmp, td)) { mtx_unlock(&np->n_mtx); error = EINTR; goto done; } } } mtx_unlock(&np->n_mtx); } else BO_UNLOCK(bo); if (NFSHASPNFS(nmp)) { nfscl_layoutcommit(vp, td); /* * Invalidate the attribute cache, since writes to a DS * won't update the size attribute. */ mtx_lock(&np->n_mtx); np->n_attrstamp = 0; } else mtx_lock(&np->n_mtx); if (np->n_flag & NWRITEERR) { error = np->n_error; np->n_flag &= ~NWRITEERR; } if (commit && bo->bo_dirty.bv_cnt == 0 && bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0) np->n_flag &= ~NMODIFIED; mtx_unlock(&np->n_mtx); done: if (bvec != NULL && bvec != bvec_on_stack) free(bvec, M_TEMP); if (error == 0 && commit != 0 && waitfor == MNT_WAIT && (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 || np->n_directio_asyncwr != 0)) { if (trycnt++ < 5) { /* try, try again... */ passone = 1; wcred = NULL; bvec = NULL; bvecsize = 0; goto again; } vn_printf(vp, "ncl_flush failed"); error = called_from_renewthread != 0 ? EIO : EBUSY; } return (error); } /* * NFS advisory byte-level locks. */ static int nfs_advlock(struct vop_advlock_args *ap) { struct vnode *vp = ap->a_vp; struct ucred *cred; struct nfsnode *np = VTONFS(ap->a_vp); struct proc *p = (struct proc *)ap->a_id; struct thread *td = curthread; /* XXX */ struct vattr va; int ret, error = EOPNOTSUPP; u_quad_t size; if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) { if (vp->v_type != VREG) return (EINVAL); if ((ap->a_flags & F_POSIX) != 0) cred = p->p_ucred; else cred = td->td_ucred; NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { NFSVOPUNLOCK(vp, 0); return (EBADF); } /* * If this is unlocking a write locked region, flush and * commit them before unlocking. This is required by * RFC3530 Sec. 9.3.2. */ if (ap->a_op == F_UNLCK && nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id, ap->a_flags)) (void) ncl_flush(vp, MNT_WAIT, td, 1, 0); /* * Loop around doing the lock op, while a blocking lock * must wait for the lock op to succeed. */ do { ret = nfsrpc_advlock(vp, np->n_size, ap->a_op, ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags); if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && ap->a_op == F_SETLK) { NFSVOPUNLOCK(vp, 0); error = nfs_catnap(PZERO | PCATCH, ret, "ncladvl"); if (error) return (EINTR); NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { NFSVOPUNLOCK(vp, 0); return (EBADF); } } } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) && ap->a_op == F_SETLK); if (ret == NFSERR_DENIED) { NFSVOPUNLOCK(vp, 0); return (EAGAIN); } else if (ret == EINVAL || ret == EBADF || ret == EINTR) { NFSVOPUNLOCK(vp, 0); return (ret); } else if (ret != 0) { NFSVOPUNLOCK(vp, 0); return (EACCES); } /* * Now, if we just got a lock, invalidate data in the buffer * cache, as required, so that the coherency conforms with * RFC3530 Sec. 9.3.2. */ if (ap->a_op == F_SETLK) { if ((np->n_flag & NMODIFIED) == 0) { np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); ret = VOP_GETATTR(vp, &va, cred); } if ((np->n_flag & NMODIFIED) || ret || np->n_change != va.va_filerev) { (void) ncl_vinvalbuf(vp, V_SAVE, td, 1); if ((vp->v_iflag & VI_DOOMED) != 0) { NFSVOPUNLOCK(vp, 0); return (EBADF); } np->n_attrstamp = 0; KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp); ret = VOP_GETATTR(vp, &va, cred); if (!ret) { np->n_mtime = va.va_mtime; np->n_change = va.va_filerev; } } /* Mark that a file lock has been acquired. */ mtx_lock(&np->n_mtx); np->n_flag |= NHASBEENLOCKED; mtx_unlock(&np->n_mtx); } NFSVOPUNLOCK(vp, 0); return (0); } else if (!NFS_ISV4(vp)) { error = NFSVOPLOCK(vp, LK_SHARED); if (error) return (error); if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { size = VTONFS(vp)->n_size; NFSVOPUNLOCK(vp, 0); error = lf_advlock(ap, &(vp->v_lockf), size); } else { if (nfs_advlock_p != NULL) error = nfs_advlock_p(ap); else { NFSVOPUNLOCK(vp, 0); error = ENOLCK; } } if (error == 0 && ap->a_op == F_SETLK) { error = NFSVOPLOCK(vp, LK_SHARED); if (error == 0) { /* Mark that a file lock has been acquired. */ mtx_lock(&np->n_mtx); np->n_flag |= NHASBEENLOCKED; mtx_unlock(&np->n_mtx); NFSVOPUNLOCK(vp, 0); } } } return (error); } /* * NFS advisory byte-level locks. */ static int nfs_advlockasync(struct vop_advlockasync_args *ap) { struct vnode *vp = ap->a_vp; u_quad_t size; int error; if (NFS_ISV4(vp)) return (EOPNOTSUPP); error = NFSVOPLOCK(vp, LK_SHARED); if (error) return (error); if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) { size = VTONFS(vp)->n_size; NFSVOPUNLOCK(vp, 0); error = lf_advlockasync(ap, &(vp->v_lockf), size); } else { NFSVOPUNLOCK(vp, 0); error = EOPNOTSUPP; } return (error); } /* * Print out the contents of an nfsnode. */ static int nfs_print(struct vop_print_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid, (uintmax_t)np->n_vattr.na_fsid); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } /* * This is the "real" nfs::bwrite(struct buf*). * We set B_CACHE if this is a VMIO buffer. */ int ncl_writebp(struct buf *bp, int force __unused, struct thread *td) { int oldflags, rtval; BUF_ASSERT_HELD(bp); if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } oldflags = bp->b_flags; bp->b_flags |= B_CACHE; /* * Undirty the bp. We will redirty it later if the I/O fails. */ bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_iocmd = BIO_WRITE; bufobj_wref(bp->b_bufobj); curthread->td_ru.ru_oublock++; /* * Note: to avoid loopback deadlocks, we do not * assign b_runningbufspace. */ vfs_busy_pages(bp, 1); BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); if ((oldflags & B_ASYNC) != 0) return (0); rtval = bufwait(bp); if (oldflags & B_DELWRI) reassignbuf(bp); brelse(bp); return (rtval); } /* * nfs special file access vnode op. * Essentially just get vattr and then imitate iaccess() since the device is * local to the client. */ static int nfsspec_access(struct vop_access_args *ap) { struct vattr *vap; struct ucred *cred = ap->a_cred; struct vnode *vp = ap->a_vp; accmode_t accmode = ap->a_accmode; struct vattr vattr; int error; /* * Disallow write attempts on filesystems mounted read-only; * unless the file is a socket, fifo, or a block or character * device resident on the filesystem. */ if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) { switch (vp->v_type) { case VREG: case VDIR: case VLNK: return (EROFS); default: break; } } vap = &vattr; error = VOP_GETATTR(vp, vap, cred); if (error) goto out; error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid, accmode, cred, NULL); out: return error; } /* * Read wrapper for fifos. */ static int nfsfifo_read(struct vop_read_args *ap) { struct nfsnode *np = VTONFS(ap->a_vp); int error; /* * Set access flag. */ mtx_lock(&np->n_mtx); np->n_flag |= NACC; vfs_timestamp(&np->n_atim); mtx_unlock(&np->n_mtx); error = fifo_specops.vop_read(ap); return error; } /* * Write wrapper for fifos. */ static int nfsfifo_write(struct vop_write_args *ap) { struct nfsnode *np = VTONFS(ap->a_vp); /* * Set update flag. */ mtx_lock(&np->n_mtx); np->n_flag |= NUPD; vfs_timestamp(&np->n_mtim); mtx_unlock(&np->n_mtx); return(fifo_specops.vop_write(ap)); } /* * Close wrapper for fifos. * * Update the times on the nfsnode then do fifo close. */ static int nfsfifo_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np = VTONFS(vp); struct vattr vattr; struct timespec ts; mtx_lock(&np->n_mtx); if (np->n_flag & (NACC | NUPD)) { vfs_timestamp(&ts); if (np->n_flag & NACC) np->n_atim = ts; if (np->n_flag & NUPD) np->n_mtim = ts; np->n_flag |= NCHG; if (vrefcnt(vp) == 1 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { VATTR_NULL(&vattr); if (np->n_flag & NACC) vattr.va_atime = np->n_atim; if (np->n_flag & NUPD) vattr.va_mtime = np->n_mtim; mtx_unlock(&np->n_mtx); (void)VOP_SETATTR(vp, &vattr, ap->a_cred); goto out; } } mtx_unlock(&np->n_mtx); out: return (fifo_specops.vop_close(ap)); } /* * Just call ncl_writebp() with the force argument set to 1. * * NOTE: B_DONE may or may not be set in a_bp on call. */ static int nfs_bwrite(struct buf *bp) { return (ncl_writebp(bp, 1, curthread)); } struct buf_ops buf_ops_newnfs = { .bop_name = "buf_ops_nfs", .bop_write = nfs_bwrite, .bop_strategy = bufstrategy, .bop_sync = bufsync, .bop_bdflush = bufbdflush, }; static int nfs_getacl(struct vop_getacl_args *ap) { int error; if (ap->a_type != ACL_TYPE_NFS4) return (EOPNOTSUPP); error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, NULL); if (error > NFSERR_STALE) { (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); error = EPERM; } return (error); } static int nfs_setacl(struct vop_setacl_args *ap) { int error; if (ap->a_type != ACL_TYPE_NFS4) return (EOPNOTSUPP); error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp, NULL); if (error > NFSERR_STALE) { (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0); error = EPERM; } return (error); } static int nfs_set_text(struct vop_set_text_args *ap) { struct vnode *vp = ap->a_vp; struct nfsnode *np; /* * If the text file has been mmap'd, flush any dirty pages to the * buffer cache and then... * Make sure all writes are pushed to the NFS server. If this is not * done, the modify time of the file can change while the text * file is being executed. This will cause the process that is * executing the text file to be terminated. */ if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC); VM_OBJECT_WUNLOCK(vp->v_object); } /* Now, flush the buffer cache. */ ncl_flush(vp, MNT_WAIT, curthread, 0, 0); /* And, finally, make sure that n_mtime is up to date. */ np = VTONFS(vp); mtx_lock(&np->n_mtx); np->n_mtime = np->n_vattr.na_mtime; mtx_unlock(&np->n_mtx); vp->v_vflag |= VV_TEXT; return (0); } /* * Return POSIX pathconf information applicable to nfs filesystems. */ static int nfs_pathconf(struct vop_pathconf_args *ap) { struct nfsv3_pathconf pc; struct nfsvattr nfsva; struct vnode *vp = ap->a_vp; struct thread *td = curthread; int attrflag, error; if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX || ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED || ap->a_name == _PC_NO_TRUNC)) || (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) { /* * Since only the above 4 a_names are returned by the NFSv3 * Pathconf RPC, there is no point in doing it for others. * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can * be used for _PC_NFS4_ACL as well. */ error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva, &attrflag, NULL); if (attrflag != 0) (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); if (error != 0) return (error); } else { /* * For NFSv2 (or NFSv3 when not one of the above 4 a_names), * just fake them. */ pc.pc_linkmax = LINK_MAX; pc.pc_namemax = NFS_MAXNAMLEN; pc.pc_notrunc = 1; pc.pc_chownrestricted = 1; pc.pc_caseinsensitive = 0; pc.pc_casepreserving = 1; error = 0; } switch (ap->a_name) { case _PC_LINK_MAX: *ap->a_retval = pc.pc_linkmax; break; case _PC_NAME_MAX: *ap->a_retval = pc.pc_namemax; break; case _PC_CHOWN_RESTRICTED: *ap->a_retval = pc.pc_chownrestricted; break; case _PC_NO_TRUNC: *ap->a_retval = pc.pc_notrunc; break; case _PC_ACL_EXTENDED: *ap->a_retval = 0; break; case _PC_ACL_NFS4: if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 && NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL)) *ap->a_retval = 1; else *ap->a_retval = 0; break; case _PC_ACL_PATH_MAX: if (NFS_ISV4(vp)) *ap->a_retval = ACL_MAX_ENTRIES; else *ap->a_retval = 3; break; case _PC_MAC_PRESENT: *ap->a_retval = 0; break; case _PC_PRIO_IO: *ap->a_retval = 0; break; case _PC_SYNC_IO: *ap->a_retval = 0; break; case _PC_ALLOC_SIZE_MIN: *ap->a_retval = vp->v_mount->mnt_stat.f_bsize; break; case _PC_FILESIZEBITS: if (NFS_ISV34(vp)) *ap->a_retval = 64; else *ap->a_retval = 32; break; case _PC_REC_INCR_XFER_SIZE: *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_MAX_XFER_SIZE: *ap->a_retval = -1; /* means ``unlimited'' */ break; case _PC_REC_MIN_XFER_SIZE: *ap->a_retval = vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_XFER_ALIGN: *ap->a_retval = PAGE_SIZE; break; case _PC_SYMLINK_MAX: *ap->a_retval = NFS_MAXPATHLEN; break; default: error = vop_stdpathconf(ap); break; } return (error); }