Index: head/sys/fs/nfs/nfs_commonkrpc.c =================================================================== --- head/sys/fs/nfs/nfs_commonkrpc.c (revision 320061) +++ head/sys/fs/nfs/nfs_commonkrpc.c (revision 320062) @@ -1,1314 +1,1323 @@ /*- * 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; + 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 (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 (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_MAXPKTHDR + + sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR + sizeof (u_int32_t)) * pktscale; - rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR + + 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)) { 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) 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/nfsport.h =================================================================== --- head/sys/fs/nfs/nfsport.h (revision 320061) +++ head/sys/fs/nfs/nfsport.h (revision 320062) @@ -1,1042 +1,1042 @@ /*- * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _NFS_NFSPORT_H_ #define _NFS_NFSPORT_H_ /* * In general, I'm not fond of #includes in .h files, but this seems * to be the cleanest way to handle #include files for the ports. */ #ifdef _KERNEL #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * For Darwin, these functions should be "static" when built in a kext. * (This is always defined as nil otherwise.) */ #define APPLESTATIC #include #include #include #include #include #include #include #include #include #include #include "opt_nfs.h" #include "opt_ufs.h" /* * These types must be defined before the nfs includes. */ #define NFSSOCKADDR_T struct sockaddr * #define NFSPROC_T struct thread #define NFSDEV_T dev_t #define NFSSVCARGS nfssvc_args #define NFSACL_T struct acl /* * These should be defined as the types used for the corresponding VOP's * argument type. */ #define NFS_ACCESS_ARGS struct vop_access_args #define NFS_OPEN_ARGS struct vop_open_args #define NFS_GETATTR_ARGS struct vop_getattr_args #define NFS_LOOKUP_ARGS struct vop_lookup_args #define NFS_READDIR_ARGS struct vop_readdir_args /* * Allocate mbufs. Must succeed and never set the mbuf ptr to NULL. */ #define NFSMGET(m) do { \ MGET((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGET((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMGETHDR(m) do { \ MGETHDR((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGETHDR((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMCLGET(m, w) do { \ MGET((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGET((m), M_WAITOK, MT_DATA); \ } \ MCLGET((m), (w)); \ } while (0) #define NFSMCLGETHDR(m, w) do { \ MGETHDR((m), M_WAITOK, MT_DATA); \ while ((m) == NULL ) { \ (void) nfs_catnap(PZERO, 0, "nfsmget"); \ MGETHDR((m), M_WAITOK, MT_DATA); \ } \ } while (0) #define NFSMTOD mtod /* * Client side constant for size of a lockowner name. */ #define NFSV4CL_LOCKNAMELEN 12 /* * Type for a mutex lock. */ #define NFSMUTEX_T struct mtx #endif /* _KERNEL */ /* * NFSv4 Operation numbers. */ #define NFSV4OP_ACCESS 3 #define NFSV4OP_CLOSE 4 #define NFSV4OP_COMMIT 5 #define NFSV4OP_CREATE 6 #define NFSV4OP_DELEGPURGE 7 #define NFSV4OP_DELEGRETURN 8 #define NFSV4OP_GETATTR 9 #define NFSV4OP_GETFH 10 #define NFSV4OP_LINK 11 #define NFSV4OP_LOCK 12 #define NFSV4OP_LOCKT 13 #define NFSV4OP_LOCKU 14 #define NFSV4OP_LOOKUP 15 #define NFSV4OP_LOOKUPP 16 #define NFSV4OP_NVERIFY 17 #define NFSV4OP_OPEN 18 #define NFSV4OP_OPENATTR 19 #define NFSV4OP_OPENCONFIRM 20 #define NFSV4OP_OPENDOWNGRADE 21 #define NFSV4OP_PUTFH 22 #define NFSV4OP_PUTPUBFH 23 #define NFSV4OP_PUTROOTFH 24 #define NFSV4OP_READ 25 #define NFSV4OP_READDIR 26 #define NFSV4OP_READLINK 27 #define NFSV4OP_REMOVE 28 #define NFSV4OP_RENAME 29 #define NFSV4OP_RENEW 30 #define NFSV4OP_RESTOREFH 31 #define NFSV4OP_SAVEFH 32 #define NFSV4OP_SECINFO 33 #define NFSV4OP_SETATTR 34 #define NFSV4OP_SETCLIENTID 35 #define NFSV4OP_SETCLIENTIDCFRM 36 #define NFSV4OP_VERIFY 37 #define NFSV4OP_WRITE 38 #define NFSV4OP_RELEASELCKOWN 39 /* * Must be one greater than the last Operation#. */ #define NFSV4OP_NOPS 40 /* * Additional Ops for NFSv4.1. */ #define NFSV4OP_BACKCHANNELCTL 40 #define NFSV4OP_BINDCONNTOSESS 41 #define NFSV4OP_EXCHANGEID 42 #define NFSV4OP_CREATESESSION 43 #define NFSV4OP_DESTROYSESSION 44 #define NFSV4OP_FREESTATEID 45 #define NFSV4OP_GETDIRDELEG 46 #define NFSV4OP_GETDEVINFO 47 #define NFSV4OP_GETDEVLIST 48 #define NFSV4OP_LAYOUTCOMMIT 49 #define NFSV4OP_LAYOUTGET 50 #define NFSV4OP_LAYOUTRETURN 51 #define NFSV4OP_SECINFONONAME 52 #define NFSV4OP_SEQUENCE 53 #define NFSV4OP_SETSSV 54 #define NFSV4OP_TESTSTATEID 55 #define NFSV4OP_WANTDELEG 56 #define NFSV4OP_DESTROYCLIENTID 57 #define NFSV4OP_RECLAIMCOMPL 58 /* * Must be one more than last op#. * NFSv4.2 isn't implemented yet, but define the op# limit for it. */ #define NFSV41_NOPS 59 #define NFSV42_NOPS 72 /* Quirky case if the illegal op code */ #define NFSV4OP_OPILLEGAL 10044 /* * Fake NFSV4OP_xxx used for nfsstat. Start at NFSV42_NOPS. */ #define NFSV4OP_SYMLINK (NFSV42_NOPS) #define NFSV4OP_MKDIR (NFSV42_NOPS + 1) #define NFSV4OP_RMDIR (NFSV42_NOPS + 2) #define NFSV4OP_READDIRPLUS (NFSV42_NOPS + 3) #define NFSV4OP_MKNOD (NFSV42_NOPS + 4) #define NFSV4OP_FSSTAT (NFSV42_NOPS + 5) #define NFSV4OP_FSINFO (NFSV42_NOPS + 6) #define NFSV4OP_PATHCONF (NFSV42_NOPS + 7) #define NFSV4OP_V3CREATE (NFSV42_NOPS + 8) /* * This is the count of the fake operations listed above. */ #define NFSV4OP_FAKENOPS 9 /* * and the Callback OPs */ #define NFSV4OP_CBGETATTR 3 #define NFSV4OP_CBRECALL 4 /* * Must be one greater than the last Callback Operation# for NFSv4.0. */ #define NFSV4OP_CBNOPS 5 /* * Additional Callback Ops for NFSv4.1 only. */ #define NFSV4OP_CBLAYOUTRECALL 5 #define NFSV4OP_CBNOTIFY 6 #define NFSV4OP_CBPUSHDELEG 7 #define NFSV4OP_CBRECALLANY 8 #define NFSV4OP_CBRECALLOBJAVAIL 9 #define NFSV4OP_CBRECALLSLOT 10 #define NFSV4OP_CBSEQUENCE 11 #define NFSV4OP_CBWANTCANCELLED 12 #define NFSV4OP_CBNOTIFYLOCK 13 #define NFSV4OP_CBNOTIFYDEVID 14 #define NFSV41_CBNOPS 15 #define NFSV42_CBNOPS 16 /* * The lower numbers -> 21 are used by NFSv2 and v3. These define higher * numbers used by NFSv4. * NFS_V3NPROCS is one greater than the last V3 op and NFS_NPROCS is * one greater than the last number. */ #ifndef NFS_V3NPROCS #define NFS_V3NPROCS 22 #define NFSPROC_LOOKUPP 22 #define NFSPROC_SETCLIENTID 23 #define NFSPROC_SETCLIENTIDCFRM 24 #define NFSPROC_LOCK 25 #define NFSPROC_LOCKU 26 #define NFSPROC_OPEN 27 #define NFSPROC_CLOSE 28 #define NFSPROC_OPENCONFIRM 29 #define NFSPROC_LOCKT 30 #define NFSPROC_OPENDOWNGRADE 31 #define NFSPROC_RENEW 32 #define NFSPROC_PUTROOTFH 33 #define NFSPROC_RELEASELCKOWN 34 #define NFSPROC_DELEGRETURN 35 #define NFSPROC_RETDELEGREMOVE 36 #define NFSPROC_RETDELEGRENAME1 37 #define NFSPROC_RETDELEGRENAME2 38 #define NFSPROC_GETACL 39 #define NFSPROC_SETACL 40 /* * Must be defined as one higher than the last Proc# above. */ #define NFSV4_NPROCS 41 /* Additional procedures for NFSv4.1. */ #define NFSPROC_EXCHANGEID 41 #define NFSPROC_CREATESESSION 42 #define NFSPROC_DESTROYSESSION 43 #define NFSPROC_DESTROYCLIENT 44 #define NFSPROC_FREESTATEID 45 #define NFSPROC_LAYOUTGET 46 #define NFSPROC_GETDEVICEINFO 47 #define NFSPROC_LAYOUTCOMMIT 48 #define NFSPROC_LAYOUTRETURN 49 #define NFSPROC_RECLAIMCOMPL 50 #define NFSPROC_WRITEDS 51 #define NFSPROC_READDS 52 #define NFSPROC_COMMITDS 53 /* * Must be defined as one higher than the last NFSv4.1 Proc# above. */ #define NFSV41_NPROCS 54 #endif /* NFS_V3NPROCS */ /* * New stats structure. * The vers field will be set to NFSSTATS_V1 by the caller. */ #define NFSSTATS_V1 1 struct nfsstatsv1 { int vers; /* Set to version requested by caller. */ uint64_t attrcache_hits; uint64_t attrcache_misses; uint64_t lookupcache_hits; uint64_t lookupcache_misses; uint64_t direofcache_hits; uint64_t direofcache_misses; uint64_t accesscache_hits; uint64_t accesscache_misses; uint64_t biocache_reads; uint64_t read_bios; uint64_t read_physios; uint64_t biocache_writes; uint64_t write_bios; uint64_t write_physios; uint64_t biocache_readlinks; uint64_t readlink_bios; uint64_t biocache_readdirs; uint64_t readdir_bios; uint64_t rpccnt[NFSV41_NPROCS + 15]; uint64_t rpcretries; uint64_t srvrpccnt[NFSV42_NOPS + NFSV4OP_FAKENOPS]; uint64_t srvrpc_errs; uint64_t srv_errs; uint64_t rpcrequests; uint64_t rpctimeouts; uint64_t rpcunexpected; uint64_t rpcinvalid; uint64_t srvcache_inproghits; uint64_t srvcache_idemdonehits; uint64_t srvcache_nonidemdonehits; uint64_t srvcache_misses; uint64_t srvcache_tcppeak; int srvcache_size; /* Updated by atomic_xx_int(). */ uint64_t srvclients; uint64_t srvopenowners; uint64_t srvopens; uint64_t srvlockowners; uint64_t srvlocks; uint64_t srvdelegates; uint64_t cbrpccnt[NFSV42_CBNOPS]; uint64_t clopenowners; uint64_t clopens; uint64_t cllockowners; uint64_t cllocks; uint64_t cldelegates; uint64_t cllocalopenowners; uint64_t cllocalopens; uint64_t cllocallockowners; uint64_t cllocallocks; uint64_t srvstartcnt; uint64_t srvdonecnt; uint64_t srvbytes[NFSV42_NOPS + NFSV4OP_FAKENOPS]; uint64_t srvops[NFSV42_NOPS + NFSV4OP_FAKENOPS]; struct bintime srvduration[NFSV42_NOPS + NFSV4OP_FAKENOPS]; struct bintime busyfrom; struct bintime busytime; }; /* * Old stats structure. */ struct ext_nfsstats { int attrcache_hits; int attrcache_misses; int lookupcache_hits; int lookupcache_misses; int direofcache_hits; int direofcache_misses; int accesscache_hits; int accesscache_misses; int biocache_reads; int read_bios; int read_physios; int biocache_writes; int write_bios; int write_physios; int biocache_readlinks; int readlink_bios; int biocache_readdirs; int readdir_bios; int rpccnt[NFSV4_NPROCS]; int rpcretries; int srvrpccnt[NFSV4OP_NOPS + NFSV4OP_FAKENOPS]; int srvrpc_errs; int srv_errs; int rpcrequests; int rpctimeouts; int rpcunexpected; int rpcinvalid; int srvcache_inproghits; int srvcache_idemdonehits; int srvcache_nonidemdonehits; int srvcache_misses; int srvcache_tcppeak; int srvcache_size; int srvclients; int srvopenowners; int srvopens; int srvlockowners; int srvlocks; int srvdelegates; int cbrpccnt[NFSV4OP_CBNOPS]; int clopenowners; int clopens; int cllockowners; int cllocks; int cldelegates; int cllocalopenowners; int cllocalopens; int cllocallockowners; int cllocallocks; }; #ifdef _KERNEL /* * Define NFS_NPROCS as NFSV4_NPROCS for the experimental kernel code. */ #ifndef NFS_NPROCS #define NFS_NPROCS NFSV4_NPROCS #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Just to keep nfs_var.h happy. */ struct nfs_vattr { int junk; }; struct nfsvattr { struct vattr na_vattr; nfsattrbit_t na_suppattr; u_int32_t na_mntonfileno; u_int64_t na_filesid[2]; }; #define na_type na_vattr.va_type #define na_mode na_vattr.va_mode #define na_nlink na_vattr.va_nlink #define na_uid na_vattr.va_uid #define na_gid na_vattr.va_gid #define na_fsid na_vattr.va_fsid #define na_fileid na_vattr.va_fileid #define na_size na_vattr.va_size #define na_blocksize na_vattr.va_blocksize #define na_atime na_vattr.va_atime #define na_mtime na_vattr.va_mtime #define na_ctime na_vattr.va_ctime #define na_gen na_vattr.va_gen #define na_flags na_vattr.va_flags #define na_rdev na_vattr.va_rdev #define na_bytes na_vattr.va_bytes #define na_filerev na_vattr.va_filerev #define na_vaflags na_vattr.va_vaflags #include /* * This is the header structure used for the lists, etc. (It has the * above record in it. */ struct nfsrv_stablefirst { LIST_HEAD(, nfsrv_stable) nsf_head; /* Head of nfsrv_stable list */ time_t nsf_eograce; /* Time grace period ends */ time_t *nsf_bootvals; /* Previous boottime values */ struct file *nsf_fp; /* File table pointer */ u_char nsf_flags; /* NFSNSF_ flags */ struct nfsf_rec nsf_rec; /* and above first record */ }; #define nsf_lease nsf_rec.lease #define nsf_numboots nsf_rec.numboots /* NFSNSF_xxx flags */ #define NFSNSF_UPDATEDONE 0x01 #define NFSNSF_GRACEOVER 0x02 #define NFSNSF_NEEDLOCK 0x04 #define NFSNSF_EXPIREDCLIENT 0x08 #define NFSNSF_NOOPENS 0x10 #define NFSNSF_OK 0x20 /* * Maximum number of boot times allowed in record. Although there is * really no need for a fixed upper bound, this serves as a sanity check * for a corrupted file. */ #define NFSNSF_MAXNUMBOOTS 10000 /* * This structure defines the other records in the file. The * nst_client array is actually the size of the client string name. */ struct nfst_rec { u_int16_t len; u_char flag; u_char client[1]; }; /* and the values for flag */ #define NFSNST_NEWSTATE 0x1 #define NFSNST_REVOKE 0x2 #define NFSNST_GOTSTATE 0x4 /* * This structure is linked onto nfsrv_stablefirst for the duration of * reclaim. */ struct nfsrv_stable { LIST_ENTRY(nfsrv_stable) nst_list; struct nfsclient *nst_clp; struct nfst_rec nst_rec; }; #define nst_timestamp nst_rec.timestamp #define nst_len nst_rec.len #define nst_flag nst_rec.flag #define nst_client nst_rec.client /* * At some point the server will run out of kernel storage for * state structures. For FreeBSD5.2, this results in a panic * kmem_map is full. It happens at well over 1000000 opens plus * locks on a PIII-800 with 256Mbytes, so that is where I've set * the limit. If your server panics due to too many opens/locks, * decrease the size of NFSRV_V4STATELIMIT. If you find the server * returning NFS4ERR_RESOURCE a lot and have lots of memory, try * increasing it. */ #define NFSRV_V4STATELIMIT 500000 /* Max # of Opens + Locks */ /* * The type required differs with BSDen (just the second arg). */ void nfsrvd_rcv(struct socket *, void *, int); /* * Macros for handling socket addresses. (Hopefully this makes the code * more portable, since I've noticed some 'BSD don't have sockaddrs in * mbufs any more.) */ #define NFSSOCKADDR(a, t) ((t)(a)) #define NFSSOCKADDRALLOC(a) \ do { \ MALLOC((a), struct sockaddr *, sizeof (struct sockaddr), \ M_SONAME, M_WAITOK); \ NFSBZERO((a), sizeof (struct sockaddr)); \ } while (0) #define NFSSOCKADDRSIZE(a, s) ((a)->sa_len = (s)) #define NFSSOCKADDRFREE(a) \ do { \ if (a) \ FREE((caddr_t)(a), M_SONAME); \ } while (0) /* * These should be defined as a process or thread structure, as required * for signal handling, etc. */ #define NFSNEWCRED(c) (crdup(c)) #define NFSPROCCRED(p) ((p)->td_ucred) #define NFSFREECRED(c) (crfree(c)) #define NFSUIOPROC(u, p) ((u)->uio_td = NULL) #define NFSPROCP(p) ((p)->td_proc) /* * Define these so that cn_hash and its length is ignored. */ #define NFSCNHASHZERO(c) #define NFSCNHASH(c, v) #define NCHNAMLEN 9999999 /* * These macros are defined to initialize and set the timer routine. */ #define NFS_TIMERINIT \ newnfs_timer(NULL) /* * Handle SMP stuff: */ #define NFSSTATESPINLOCK extern struct mtx nfs_state_mutex #define NFSLOCKSTATE() mtx_lock(&nfs_state_mutex) #define NFSUNLOCKSTATE() mtx_unlock(&nfs_state_mutex) #define NFSSTATEMUTEXPTR (&nfs_state_mutex) #define NFSREQSPINLOCK extern struct mtx nfs_req_mutex #define NFSLOCKREQ() mtx_lock(&nfs_req_mutex) #define NFSUNLOCKREQ() mtx_unlock(&nfs_req_mutex) #define NFSSOCKMUTEX extern struct mtx nfs_slock_mutex #define NFSSOCKMUTEXPTR (&nfs_slock_mutex) #define NFSLOCKSOCK() mtx_lock(&nfs_slock_mutex) #define NFSUNLOCKSOCK() mtx_unlock(&nfs_slock_mutex) #define NFSNAMEIDMUTEX extern struct mtx nfs_nameid_mutex #define NFSLOCKNAMEID() mtx_lock(&nfs_nameid_mutex) #define NFSUNLOCKNAMEID() mtx_unlock(&nfs_nameid_mutex) #define NFSNAMEIDREQUIRED() mtx_assert(&nfs_nameid_mutex, MA_OWNED) #define NFSCLSTATEMUTEX extern struct mtx nfs_clstate_mutex #define NFSCLSTATEMUTEXPTR (&nfs_clstate_mutex) #define NFSLOCKCLSTATE() mtx_lock(&nfs_clstate_mutex) #define NFSUNLOCKCLSTATE() mtx_unlock(&nfs_clstate_mutex) #define NFSDLOCKMUTEX extern struct mtx newnfsd_mtx #define NFSDLOCKMUTEXPTR (&newnfsd_mtx) #define NFSD_LOCK() mtx_lock(&newnfsd_mtx) #define NFSD_UNLOCK() mtx_unlock(&newnfsd_mtx) #define NFSD_LOCK_ASSERT() mtx_assert(&newnfsd_mtx, MA_OWNED) #define NFSD_UNLOCK_ASSERT() mtx_assert(&newnfsd_mtx, MA_NOTOWNED) #define NFSV4ROOTLOCKMUTEX extern struct mtx nfs_v4root_mutex #define NFSV4ROOTLOCKMUTEXPTR (&nfs_v4root_mutex) #define NFSLOCKV4ROOTMUTEX() mtx_lock(&nfs_v4root_mutex) #define NFSUNLOCKV4ROOTMUTEX() mtx_unlock(&nfs_v4root_mutex) #define NFSLOCKNODE(n) mtx_lock(&((n)->n_mtx)) #define NFSUNLOCKNODE(n) mtx_unlock(&((n)->n_mtx)) #define NFSLOCKMNT(m) mtx_lock(&((m)->nm_mtx)) #define NFSUNLOCKMNT(m) mtx_unlock(&((m)->nm_mtx)) #define NFSLOCKREQUEST(r) mtx_lock(&((r)->r_mtx)) #define NFSUNLOCKREQUEST(r) mtx_unlock(&((r)->r_mtx)) #define NFSPROCLISTLOCK() sx_slock(&allproc_lock) #define NFSPROCLISTUNLOCK() sx_sunlock(&allproc_lock) #define NFSLOCKSOCKREQ(r) mtx_lock(&((r)->nr_mtx)) #define NFSUNLOCKSOCKREQ(r) mtx_unlock(&((r)->nr_mtx)) #define NFSLOCKDS(d) mtx_lock(&((d)->nfsclds_mtx)) #define NFSUNLOCKDS(d) mtx_unlock(&((d)->nfsclds_mtx)) #define NFSSESSIONMUTEXPTR(s) (&((s)->mtx)) #define NFSLOCKSESSION(s) mtx_lock(&((s)->mtx)) #define NFSUNLOCKSESSION(s) mtx_unlock(&((s)->mtx)) /* * Use these macros to initialize/free a mutex. */ #define NFSINITSOCKMUTEX(m) mtx_init((m), "nfssock", NULL, MTX_DEF) #define NFSFREEMUTEX(m) mtx_destroy((m)) int nfsmsleep(void *, void *, int, const char *, struct timespec *); /* * And weird vm stuff in the nfs server. */ #define PDIRUNLOCK 0x0 #define MAX_COMMIT_COUNT (1024 * 1024) /* * Define these to handle the type of va_rdev. */ #define NFSMAKEDEV(m, n) makedev((m), (n)) #define NFSMAJOR(d) major(d) #define NFSMINOR(d) minor(d) /* * The vnode tag for nfsv4root. */ #define VT_NFSV4ROOT "nfsv4root" /* * Define whatever it takes to do a vn_rdwr(). */ #define NFSD_RDWR(r, v, b, l, o, s, i, c, a, p) \ vn_rdwr((r), (v), (b), (l), (o), (s), (i), (c), NULL, (a), (p)) /* * Macros for handling memory for different BSDen. * NFSBCOPY(src, dst, len) - copies len bytes, non-overlapping * NFSOVBCOPY(src, dst, len) - ditto, but data areas might overlap * NFSBCMP(cp1, cp2, len) - compare len bytes, return 0 if same * NFSBZERO(cp, len) - set len bytes to 0x0 */ #define NFSBCOPY(s, d, l) bcopy((s), (d), (l)) #define NFSOVBCOPY(s, d, l) ovbcopy((s), (d), (l)) #define NFSBCMP(s, d, l) bcmp((s), (d), (l)) #define NFSBZERO(s, l) bzero((s), (l)) /* * Some queue.h files don't have these dfined in them. */ #define LIST_END(head) NULL #define SLIST_END(head) NULL #define TAILQ_END(head) NULL /* * This must be defined to be a global variable that increments once * per second, but never stops or goes backwards, even when a "date" * command changes the TOD clock. It is used for delta times for * leases, etc. */ #define NFSD_MONOSEC time_uptime /* * Declare the malloc types. */ MALLOC_DECLARE(M_NEWNFSRVCACHE); MALLOC_DECLARE(M_NEWNFSDCLIENT); MALLOC_DECLARE(M_NEWNFSDSTATE); MALLOC_DECLARE(M_NEWNFSDLOCK); MALLOC_DECLARE(M_NEWNFSDLOCKFILE); MALLOC_DECLARE(M_NEWNFSSTRING); MALLOC_DECLARE(M_NEWNFSUSERGROUP); MALLOC_DECLARE(M_NEWNFSDREQ); MALLOC_DECLARE(M_NEWNFSFH); MALLOC_DECLARE(M_NEWNFSCLOWNER); MALLOC_DECLARE(M_NEWNFSCLOPEN); MALLOC_DECLARE(M_NEWNFSCLDELEG); MALLOC_DECLARE(M_NEWNFSCLCLIENT); MALLOC_DECLARE(M_NEWNFSCLLOCKOWNER); MALLOC_DECLARE(M_NEWNFSCLLOCK); MALLOC_DECLARE(M_NEWNFSDIROFF); MALLOC_DECLARE(M_NEWNFSV4NODE); MALLOC_DECLARE(M_NEWNFSDIRECTIO); MALLOC_DECLARE(M_NEWNFSMNT); MALLOC_DECLARE(M_NEWNFSDROLLBACK); MALLOC_DECLARE(M_NEWNFSLAYOUT); MALLOC_DECLARE(M_NEWNFSFLAYOUT); MALLOC_DECLARE(M_NEWNFSDEVINFO); MALLOC_DECLARE(M_NEWNFSSOCKREQ); MALLOC_DECLARE(M_NEWNFSCLDS); MALLOC_DECLARE(M_NEWNFSLAYRECALL); MALLOC_DECLARE(M_NEWNFSDSESSION); #define M_NFSRVCACHE M_NEWNFSRVCACHE #define M_NFSDCLIENT M_NEWNFSDCLIENT #define M_NFSDSTATE M_NEWNFSDSTATE #define M_NFSDLOCK M_NEWNFSDLOCK #define M_NFSDLOCKFILE M_NEWNFSDLOCKFILE #define M_NFSSTRING M_NEWNFSSTRING #define M_NFSUSERGROUP M_NEWNFSUSERGROUP #define M_NFSDREQ M_NEWNFSDREQ #define M_NFSFH M_NEWNFSFH #define M_NFSCLOWNER M_NEWNFSCLOWNER #define M_NFSCLOPEN M_NEWNFSCLOPEN #define M_NFSCLDELEG M_NEWNFSCLDELEG #define M_NFSCLCLIENT M_NEWNFSCLCLIENT #define M_NFSCLLOCKOWNER M_NEWNFSCLLOCKOWNER #define M_NFSCLLOCK M_NEWNFSCLLOCK #define M_NFSDIROFF M_NEWNFSDIROFF #define M_NFSV4NODE M_NEWNFSV4NODE #define M_NFSDIRECTIO M_NEWNFSDIRECTIO #define M_NFSDROLLBACK M_NEWNFSDROLLBACK #define M_NFSLAYOUT M_NEWNFSLAYOUT #define M_NFSFLAYOUT M_NEWNFSFLAYOUT #define M_NFSDEVINFO M_NEWNFSDEVINFO #define M_NFSSOCKREQ M_NEWNFSSOCKREQ #define M_NFSCLDS M_NEWNFSCLDS #define M_NFSLAYRECALL M_NEWNFSLAYRECALL #define M_NFSDSESSION M_NEWNFSDSESSION #define NFSINT_SIGMASK(set) \ (SIGISMEMBER(set, SIGINT) || SIGISMEMBER(set, SIGTERM) || \ SIGISMEMBER(set, SIGHUP) || SIGISMEMBER(set, SIGKILL) || \ SIGISMEMBER(set, SIGQUIT)) /* * Convert a quota block count to byte count. */ #define NFSQUOTABLKTOBYTE(q, b) (q) *= (b) /* * Define this as the largest file size supported. (It should probably * be available via a VFS_xxx Op, but it isn't. */ #define NFSRV_MAXFILESIZE ((u_int64_t)0x800000000000) /* * Set this macro to index() or strchr(), whichever is supported. */ #define STRCHR(s, c) strchr((s), (c)) /* * Set the n_time in the client write rpc, as required. */ #define NFSWRITERPC_SETTIME(w, n, a, v4) \ do { \ if (w) { \ mtx_lock(&((n)->n_mtx)); \ (n)->n_mtime = (a)->na_mtime; \ if (v4) \ (n)->n_change = (a)->na_filerev; \ mtx_unlock(&((n)->n_mtx)); \ } \ } while (0) /* * Fake value, just to make the client work. */ #define NFS_LATTR_NOSHRINK 1 /* * Prototypes for functions where the arguments vary for different ports. */ int nfscl_loadattrcache(struct vnode **, struct nfsvattr *, void *, void *, int, int); int newnfs_realign(struct mbuf **, int); /* * If the port runs on an SMP box that can enforce Atomic ops with low * overheads, define these as atomic increments/decrements. If not, * don't worry about it, since these are used for stats that can be * "out by one" without disastrous consequences. */ #define NFSINCRGLOBAL(a) ((a)++) /* * Assorted funky stuff to make things work under Darwin8. */ /* * These macros checks for a field in vattr being set. */ #define NFSATTRISSET(t, v, a) ((v)->a != (t)VNOVAL) #define NFSATTRISSETTIME(v, a) ((v)->a.tv_sec != VNOVAL) /* * Manipulate mount flags. */ #define NFSSTA_HASWRITEVERF 0x00040000 /* Has write verifier */ #define NFSSTA_GOTFSINFO 0x00100000 /* Got the fsinfo */ #define NFSSTA_OPENMODE 0x00200000 /* Must use correct open mode */ #define NFSSTA_NOLAYOUTCOMMIT 0x04000000 /* Don't do LayoutCommit */ #define NFSSTA_SESSPERSIST 0x08000000 /* Has a persistent session */ #define NFSSTA_TIMEO 0x10000000 /* Experiencing a timeout */ #define NFSSTA_LOCKTIMEO 0x20000000 /* Experiencing a lockd timeout */ #define NFSSTA_HASSETFSID 0x40000000 /* Has set the fsid */ #define NFSSTA_PNFS 0x80000000 /* pNFS is enabled */ #define NFSHASNFSV3(n) ((n)->nm_flag & NFSMNT_NFSV3) #define NFSHASNFSV4(n) ((n)->nm_flag & NFSMNT_NFSV4) #define NFSHASNFSV4N(n) ((n)->nm_minorvers > 0) #define NFSHASNFSV3OR4(n) ((n)->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) #define NFSHASGOTFSINFO(n) ((n)->nm_state & NFSSTA_GOTFSINFO) #define NFSHASHASSETFSID(n) ((n)->nm_state & NFSSTA_HASSETFSID) #define NFSHASSTRICT3530(n) ((n)->nm_flag & NFSMNT_STRICT3530) #define NFSHASWRITEVERF(n) ((n)->nm_state & NFSSTA_HASWRITEVERF) #define NFSHASINT(n) ((n)->nm_flag & NFSMNT_INT) #define NFSHASSOFT(n) ((n)->nm_flag & NFSMNT_SOFT) #define NFSHASINTORSOFT(n) ((n)->nm_flag & (NFSMNT_INT | NFSMNT_SOFT)) #define NFSHASDUMBTIMR(n) ((n)->nm_flag & NFSMNT_DUMBTIMR) #define NFSHASNOCONN(n) ((n)->nm_flag & NFSMNT_MNTD) #define NFSHASKERB(n) ((n)->nm_flag & NFSMNT_KERB) #define NFSHASALLGSSNAME(n) ((n)->nm_flag & NFSMNT_ALLGSSNAME) #define NFSHASINTEGRITY(n) ((n)->nm_flag & NFSMNT_INTEGRITY) #define NFSHASPRIVACY(n) ((n)->nm_flag & NFSMNT_PRIVACY) #define NFSSETWRITEVERF(n) ((n)->nm_state |= NFSSTA_HASWRITEVERF) #define NFSSETHASSETFSID(n) ((n)->nm_state |= NFSSTA_HASSETFSID) #define NFSHASPNFSOPT(n) ((n)->nm_flag & NFSMNT_PNFS) #define NFSHASNOLAYOUTCOMMIT(n) ((n)->nm_state & NFSSTA_NOLAYOUTCOMMIT) #define NFSHASSESSPERSIST(n) ((n)->nm_state & NFSSTA_SESSPERSIST) #define NFSHASPNFS(n) ((n)->nm_state & NFSSTA_PNFS) #define NFSHASOPENMODE(n) ((n)->nm_state & NFSSTA_OPENMODE) #define NFSHASONEOPENOWN(n) (((n)->nm_flag & NFSMNT_ONEOPENOWN) != 0 && \ (n)->nm_minorvers > 0) /* * Gets the stats field out of the mount structure. */ #define vfs_statfs(m) (&((m)->mnt_stat)) /* * Set boottime. */ #define NFSSETBOOTTIME(b) (getboottime(&b)) /* * The size of directory blocks in the buffer cache. * MUST BE in the range of PAGE_SIZE <= NFS_DIRBLKSIZ <= MAXBSIZE!! */ #define NFS_DIRBLKSIZ (16 * DIRBLKSIZ) /* Must be a multiple of DIRBLKSIZ */ /* * Define these macros to access mnt_flag fields. */ #define NFSMNT_RDONLY(m) ((m)->mnt_flag & MNT_RDONLY) #endif /* _KERNEL */ /* * Define a structure similar to ufs_args for use in exporting the V4 root. */ struct nfsex_args { char *fspec; struct export_args export; }; /* * These export flags should be defined, but there are no bits left. * Maybe a separate mnt_exflag field could be added or the mnt_flag * field increased to 64 bits? */ #ifndef MNT_EXSTRICTACCESS #define MNT_EXSTRICTACCESS 0x0 #endif #ifndef MNT_EXV4ONLY #define MNT_EXV4ONLY 0x0 #endif #ifdef _KERNEL /* * Define this to invalidate the attribute cache for the nfs node. */ #define NFSINVALATTRCACHE(n) ((n)->n_attrstamp = 0) /* Used for FreeBSD only */ void nfsd_mntinit(void); /* * Define these for vnode lock/unlock ops. * * These are good abstractions to macro out, so that they can be added to * later, for debugging or stats, etc. */ #define NFSVOPLOCK(v, f) vn_lock((v), (f)) #define NFSVOPUNLOCK(v, f) VOP_UNLOCK((v), (f)) #define NFSVOPISLOCKED(v) VOP_ISLOCKED((v)) /* * Define ncl_hash(). */ #define ncl_hash(f, l) (fnv_32_buf((f), (l), FNV1_32_INIT)) int newnfs_iosize(struct nfsmount *); int newnfs_vncmpf(struct vnode *, void *); #ifndef NFS_MINDIRATTRTIMO #define NFS_MINDIRATTRTIMO 3 /* VDIR attrib cache timeout in sec */ #endif #ifndef NFS_MAXDIRATTRTIMO #define NFS_MAXDIRATTRTIMO 60 #endif /* * Nfs outstanding request list element */ struct nfsreq { TAILQ_ENTRY(nfsreq) r_chain; u_int32_t r_flags; /* flags on request, see below */ struct nfsmount *r_nmp; /* Client mnt ptr */ struct mtx r_mtx; /* Mutex lock for this structure */ }; #ifndef NFS_MAXBSIZE -#define NFS_MAXBSIZE MAXBCACHEBUF +#define NFS_MAXBSIZE (maxbcachebuf) #endif /* * This macro checks to see if issuing of delegations is allowed for this * vnode. */ #ifdef VV_DISABLEDELEG #define NFSVNO_DELEGOK(v) \ ((v) == NULL || ((v)->v_vflag & VV_DISABLEDELEG) == 0) #else #define NFSVNO_DELEGOK(v) (1) #endif /* * Name used by getnewvnode() to describe filesystem, "nfs". * For performance reasons it is useful to have the same string * used in both places that call getnewvnode(). */ extern const char nfs_vnode_tag[]; #endif /* _KERNEL */ #endif /* _NFS_NFSPORT_H */ Index: head/sys/fs/nfsclient/nfs_clrpcops.c =================================================================== --- head/sys/fs/nfsclient/nfs_clrpcops.c (revision 320061) +++ head/sys/fs/nfsclient/nfs_clrpcops.c (revision 320062) @@ -1,6015 +1,6034 @@ /*- * 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("$FreeBSD$"); /* * Rpc op calls, generally called from the vnode op calls or through the * buffer cache, for NFS v2, 3 and 4. * These do not normally make any changes to vnode arguments or use * structures that might change between the VFS variants. The returned * arguments are all at the end, after the NFSPROC_T *p one. */ #ifndef APPLEKEXT #include "opt_inet6.h" #include #include SYSCTL_DECL(_vfs_nfs); static int nfsignore_eexist = 0; SYSCTL_INT(_vfs_nfs, OID_AUTO, ignore_eexist, CTLFLAG_RW, &nfsignore_eexist, 0, "NFS ignore EEXIST replies for mkdir/symlink"); /* * Global variables */ extern int nfs_numnfscbd; extern struct timeval nfsboottime; extern u_int32_t newnfs_false, newnfs_true; extern nfstype nfsv34_type[9]; extern int nfsrv_useacl; extern char nfsv4_callbackaddr[INET6_ADDRSTRLEN]; extern int nfscl_debuglevel; NFSCLSTATEMUTEX; int nfstest_outofseq = 0; int nfscl_assumeposixlocks = 1; int nfscl_enablecallb = 0; short nfsv4_cbport = NFSV4_CBPORT; int nfstest_openallsetattr = 0; #endif /* !APPLEKEXT */ #define DIRHDSIZ offsetof(struct dirent, d_name) /* * nfscl_getsameserver() can return one of three values: * NFSDSP_USETHISSESSION - Use this session for the DS. * NFSDSP_SEQTHISSESSION - Use the nfsclds_sequence field of this dsp for new * session. * NFSDSP_NOTFOUND - No matching server was found. */ enum nfsclds_state { NFSDSP_USETHISSESSION = 0, NFSDSP_SEQTHISSESSION = 1, NFSDSP_NOTFOUND = 2, }; static int nfsrpc_setattrrpc(vnode_t , struct vattr *, nfsv4stateid_t *, struct ucred *, NFSPROC_T *, struct nfsvattr *, int *, void *); static int nfsrpc_readrpc(vnode_t , struct uio *, struct ucred *, nfsv4stateid_t *, NFSPROC_T *, struct nfsvattr *, int *, void *); static int nfsrpc_writerpc(vnode_t , struct uio *, int *, int *, struct ucred *, nfsv4stateid_t *, NFSPROC_T *, struct nfsvattr *, int *, void *); static int nfsrpc_createv23(vnode_t , char *, int, struct vattr *, nfsquad_t, int, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *); static int nfsrpc_createv4(vnode_t , char *, int, struct vattr *, nfsquad_t, int, struct nfsclowner *, struct nfscldeleg **, struct ucred *, NFSPROC_T *, struct nfsvattr *, struct nfsvattr *, struct nfsfh **, int *, int *, void *, int *); static int nfsrpc_locku(struct nfsrv_descript *, struct nfsmount *, struct nfscllockowner *, u_int64_t, u_int64_t, u_int32_t, struct ucred *, NFSPROC_T *, int); static int nfsrpc_setaclrpc(vnode_t, struct ucred *, NFSPROC_T *, struct acl *, nfsv4stateid_t *, void *); static int nfsrpc_getlayout(struct nfsmount *, vnode_t, struct nfsfh *, int, uint32_t *, nfsv4stateid_t *, uint64_t, struct nfscllayout **, struct ucred *, NFSPROC_T *); static int nfsrpc_fillsa(struct nfsmount *, struct sockaddr_storage *, struct nfsclds **, NFSPROC_T *); static void nfscl_initsessionslots(struct nfsclsession *); static int nfscl_doflayoutio(vnode_t, struct uio *, int *, int *, int *, nfsv4stateid_t *, int, struct nfscldevinfo *, struct nfscllayout *, struct nfsclflayout *, uint64_t, uint64_t, struct ucred *, NFSPROC_T *); static int nfsrpc_readds(vnode_t, struct uio *, nfsv4stateid_t *, int *, struct nfsclds *, uint64_t, int, struct nfsfh *, struct ucred *, NFSPROC_T *); static int nfsrpc_writeds(vnode_t, struct uio *, int *, int *, nfsv4stateid_t *, struct nfsclds *, uint64_t, int, struct nfsfh *, int, struct ucred *, NFSPROC_T *); static enum nfsclds_state nfscl_getsameserver(struct nfsmount *, struct nfsclds *, struct nfsclds **); #ifdef notyet static int nfsrpc_commitds(vnode_t, uint64_t, int, struct nfsclds *, struct nfsfh *, struct ucred *, NFSPROC_T *, void *); #endif /* * nfs null call from vfs. */ APPLESTATIC int nfsrpc_null(vnode_t vp, struct ucred *cred, NFSPROC_T *p) { int error; struct nfsrv_descript nfsd, *nd = &nfsd; NFSCL_REQSTART(nd, NFSPROC_NULL, vp); error = nfscl_request(nd, vp, p, cred, NULL); if (nd->nd_repstat && !error) error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * nfs access rpc op. * For nfs version 3 and 4, use the access rpc to check accessibility. If file * modes are changed on the server, accesses might still fail later. */ APPLESTATIC int nfsrpc_access(vnode_t vp, int acmode, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp) { int error; u_int32_t mode, rmode; if (acmode & VREAD) mode = NFSACCESS_READ; else mode = 0; if (vnode_vtype(vp) == VDIR) { if (acmode & VWRITE) mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND | NFSACCESS_DELETE); if (acmode & VEXEC) mode |= NFSACCESS_LOOKUP; } else { if (acmode & VWRITE) mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND); if (acmode & VEXEC) mode |= NFSACCESS_EXECUTE; } /* * Now, just call nfsrpc_accessrpc() to do the actual RPC. */ error = nfsrpc_accessrpc(vp, mode, cred, p, nap, attrflagp, &rmode, NULL); /* * The NFS V3 spec does not clarify whether or not * the returned access bits can be a superset of * the ones requested, so... */ if (!error && (rmode & mode) != mode) error = EACCES; return (error); } /* * The actual rpc, separated out for Darwin. */ APPLESTATIC int nfsrpc_accessrpc(vnode_t vp, u_int32_t mode, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, u_int32_t *rmodep, void *stuff) { u_int32_t *tl; u_int32_t supported, rmode; int error; struct nfsrv_descript nfsd, *nd = &nfsd; nfsattrbit_t attrbits; *attrflagp = 0; supported = mode; NFSCL_REQSTART(nd, NFSPROC_ACCESS, vp); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(mode); if (nd->nd_flag & ND_NFSV4) { /* * And do a Getattr op. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_flag & ND_NFSV3) { error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (error) goto nfsmout; } if (!nd->nd_repstat) { if (nd->nd_flag & ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); supported = fxdr_unsigned(u_int32_t, *tl++); } else { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); } rmode = fxdr_unsigned(u_int32_t, *tl); if (nd->nd_flag & ND_NFSV4) error = nfscl_postop_attr(nd, nap, attrflagp, stuff); /* * It's not obvious what should be done about * unsupported access modes. For now, be paranoid * and clear the unsupported ones. */ rmode &= supported; *rmodep = rmode; } else error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs open rpc */ APPLESTATIC int nfsrpc_open(vnode_t vp, int amode, struct ucred *cred, NFSPROC_T *p) { struct nfsclopen *op; struct nfscldeleg *dp; struct nfsfh *nfhp; struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); u_int32_t mode, clidrev; int ret, newone, error, expireret = 0, retrycnt; /* * For NFSv4, Open Ops are only done on Regular Files. */ if (vnode_vtype(vp) != VREG) return (0); mode = 0; if (amode & FREAD) mode |= NFSV4OPEN_ACCESSREAD; if (amode & FWRITE) mode |= NFSV4OPEN_ACCESSWRITE; nfhp = np->n_fhp; retrycnt = 0; #ifdef notdef { char name[100]; int namel; namel = (np->n_v4->n4_namelen < 100) ? np->n_v4->n4_namelen : 99; bcopy(NFS4NODENAME(np->n_v4), name, namel); name[namel] = '\0'; printf("rpcopen p=0x%x name=%s",p->p_pid,name); if (nfhp->nfh_len > 0) printf(" fh=0x%x\n",nfhp->nfh_fh[12]); else printf(" fhl=0\n"); } #endif do { dp = NULL; error = nfscl_open(vp, nfhp->nfh_fh, nfhp->nfh_len, mode, 1, cred, p, NULL, &op, &newone, &ret, 1); if (error) { return (error); } if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; else clidrev = 0; if (ret == NFSCLOPEN_DOOPEN) { if (np->n_v4 != NULL) { error = nfsrpc_openrpc(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, mode, op, NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &dp, 0, 0x0, cred, p, 0, 0); if (dp != NULL) { #ifdef APPLE OSBitAndAtomic((int32_t)~NDELEGMOD, (UInt32 *)&np->n_flag); #else NFSLOCKNODE(np); np->n_flag &= ~NDELEGMOD; /* * Invalidate the attribute cache, so that * attributes that pre-date the issue of a * delegation are not cached, since the * cached attributes will remain valid while * the delegation is held. */ NFSINVALATTRCACHE(np); NFSUNLOCKNODE(np); #endif (void) nfscl_deleg(nmp->nm_mountp, op->nfso_own->nfsow_clp, nfhp->nfh_fh, nfhp->nfh_len, cred, p, &dp); } } else { error = EIO; } newnfs_copyincred(cred, &op->nfso_cred); } else if (ret == NFSCLOPEN_SETCRED) /* * This is a new local open on a delegation. It needs * to have credentials so that an open can be done * against the server during recovery. */ newnfs_copyincred(cred, &op->nfso_cred); /* * nfso_opencnt is the count of how many VOP_OPEN()s have * been done on this Open successfully and a VOP_CLOSE() * is expected for each of these. * If error is non-zero, don't increment it, since the Open * hasn't succeeded yet. */ if (!error) op->nfso_opencnt++; nfscl_openrelease(nmp, op, error, newone); if (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_BADSESSION) { (void) nfs_catnap(PZERO, error, "nfs_open"); } else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && clidrev != 0) { expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p); retrycnt++; } } while (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_BADSESSION || ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && expireret == 0 && clidrev != 0 && retrycnt < 4)); if (error && retrycnt >= 4) error = EIO; return (error); } /* * the actual open rpc */ APPLESTATIC int nfsrpc_openrpc(struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, int fhlen, u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op, u_int8_t *name, int namelen, struct nfscldeleg **dpp, int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p, int syscred, int recursed) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfscldeleg *dp, *ndp = NULL; struct nfsvattr nfsva; u_int32_t rflags, deleg; nfsattrbit_t attrbits; int error, ret, acesize, limitby; struct nfsclsession *tsep; dp = *dpp; *dpp = NULL; nfscl_reqstart(nd, NFSPROC_OPEN, nmp, nfhp, fhlen, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(op->nfso_own->nfsow_seqid); *tl++ = txdr_unsigned(mode & NFSV4OPEN_ACCESSBOTH); *tl++ = txdr_unsigned((mode >> NFSLCK_SHIFT) & NFSV4OPEN_DENYBOTH); tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; (void) nfsm_strtom(nd, op->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OPEN_NOCREATE); if (reclaim) { *tl = txdr_unsigned(NFSV4OPEN_CLAIMPREVIOUS); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(delegtype); } else { if (dp != NULL) { *tl = txdr_unsigned(NFSV4OPEN_CLAIMDELEGATECUR); NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = dp->nfsdl_stateid.seqid; *tl++ = dp->nfsdl_stateid.other[0]; *tl++ = dp->nfsdl_stateid.other[1]; *tl = dp->nfsdl_stateid.other[2]; } else { *tl = txdr_unsigned(NFSV4OPEN_CLAIMNULL); } (void) nfsm_strtom(nd, name, namelen); } NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY); (void) nfsrv_putattrbit(nd, &attrbits); if (syscred) nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); NFSCL_INCRSEQID(op->nfso_own->nfsow_seqid, nd); if (!nd->nd_repstat) { NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID + 6 * NFSX_UNSIGNED); op->nfso_stateid.seqid = *tl++; op->nfso_stateid.other[0] = *tl++; op->nfso_stateid.other[1] = *tl++; op->nfso_stateid.other[2] = *tl; rflags = fxdr_unsigned(u_int32_t, *(tl + 6)); error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); deleg = fxdr_unsigned(u_int32_t, *tl); if (deleg == NFSV4OPEN_DELEGATEREAD || deleg == NFSV4OPEN_DELEGATEWRITE) { if (!(op->nfso_own->nfsow_clp->nfsc_flags & NFSCLFLAGS_FIRSTDELEG)) op->nfso_own->nfsow_clp->nfsc_flags |= (NFSCLFLAGS_FIRSTDELEG | NFSCLFLAGS_GOTDELEG); MALLOC(ndp, struct nfscldeleg *, sizeof (struct nfscldeleg) + newfhlen, M_NFSCLDELEG, M_WAITOK); LIST_INIT(&ndp->nfsdl_owner); LIST_INIT(&ndp->nfsdl_lock); ndp->nfsdl_clp = op->nfso_own->nfsow_clp; ndp->nfsdl_fhlen = newfhlen; NFSBCOPY(newfhp, ndp->nfsdl_fh, newfhlen); newnfs_copyincred(cred, &ndp->nfsdl_cred); nfscl_lockinit(&ndp->nfsdl_rwlock); NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID + NFSX_UNSIGNED); ndp->nfsdl_stateid.seqid = *tl++; ndp->nfsdl_stateid.other[0] = *tl++; ndp->nfsdl_stateid.other[1] = *tl++; ndp->nfsdl_stateid.other[2] = *tl++; ret = fxdr_unsigned(int, *tl); if (deleg == NFSV4OPEN_DELEGATEWRITE) { ndp->nfsdl_flags = NFSCLDL_WRITE; /* * Indicates how much the file can grow. */ NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); limitby = fxdr_unsigned(int, *tl++); switch (limitby) { case NFSV4OPEN_LIMITSIZE: ndp->nfsdl_sizelimit = fxdr_hyper(tl); break; case NFSV4OPEN_LIMITBLOCKS: ndp->nfsdl_sizelimit = fxdr_unsigned(u_int64_t, *tl++); ndp->nfsdl_sizelimit *= fxdr_unsigned(u_int64_t, *tl); break; default: error = NFSERR_BADXDR; goto nfsmout; } } else { ndp->nfsdl_flags = NFSCLDL_READ; } if (ret) ndp->nfsdl_flags |= NFSCLDL_RECALL; error = nfsrv_dissectace(nd, &ndp->nfsdl_ace, &ret, &acesize, p); if (error) goto nfsmout; } else if (deleg != NFSV4OPEN_DELEGATENONE) { error = NFSERR_BADXDR; goto nfsmout; } NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, p, cred); if (error) goto nfsmout; if (ndp != NULL) { ndp->nfsdl_change = nfsva.na_filerev; ndp->nfsdl_modtime = nfsva.na_mtime; ndp->nfsdl_flags |= NFSCLDL_MODTIMESET; } if (!reclaim && (rflags & NFSV4OPEN_RESULTCONFIRM)) { do { ret = nfsrpc_openconfirm(vp, newfhp, newfhlen, op, cred, p); if (ret == NFSERR_DELAY) (void) nfs_catnap(PZERO, ret, "nfs_open"); } while (ret == NFSERR_DELAY); error = ret; } if ((rflags & NFSV4OPEN_LOCKTYPEPOSIX) || nfscl_assumeposixlocks) op->nfso_posixlock = 1; else op->nfso_posixlock = 0; /* * If the server is handing out delegations, but we didn't * get one because an OpenConfirm was required, try the * Open again, to get a delegation. This is a harmless no-op, * from a server's point of view. */ if (!reclaim && (rflags & NFSV4OPEN_RESULTCONFIRM) && (op->nfso_own->nfsow_clp->nfsc_flags & NFSCLFLAGS_GOTDELEG) && !error && dp == NULL && ndp == NULL && !recursed) { do { ret = nfsrpc_openrpc(nmp, vp, nfhp, fhlen, newfhp, newfhlen, mode, op, name, namelen, &ndp, 0, 0x0, cred, p, syscred, 1); if (ret == NFSERR_DELAY) (void) nfs_catnap(PZERO, ret, "nfs_open2"); } while (ret == NFSERR_DELAY); if (ret) { if (ndp != NULL) { FREE((caddr_t)ndp, M_NFSCLDELEG); ndp = NULL; } if (ret == NFSERR_STALECLIENTID || ret == NFSERR_STALEDONTRECOVER || ret == NFSERR_BADSESSION) error = ret; } } } if (nd->nd_repstat != 0 && error == 0) error = nd->nd_repstat; if (error == NFSERR_STALECLIENTID) nfscl_initiate_recovery(op->nfso_own->nfsow_clp); nfsmout: if (!error) *dpp = ndp; else if (ndp != NULL) FREE((caddr_t)ndp, M_NFSCLDELEG); mbuf_freem(nd->nd_mrep); return (error); } /* * open downgrade rpc */ APPLESTATIC int nfsrpc_opendowngrade(vnode_t vp, u_int32_t mode, struct nfsclopen *op, struct ucred *cred, NFSPROC_T *p) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; int error; NFSCL_REQSTART(nd, NFSPROC_OPENDOWNGRADE, vp); NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID + 3 * NFSX_UNSIGNED); if (NFSHASNFSV4N(VFSTONFS(vnode_mount(vp)))) *tl++ = 0; else *tl++ = op->nfso_stateid.seqid; *tl++ = op->nfso_stateid.other[0]; *tl++ = op->nfso_stateid.other[1]; *tl++ = op->nfso_stateid.other[2]; *tl++ = txdr_unsigned(op->nfso_own->nfsow_seqid); *tl++ = txdr_unsigned(mode & NFSV4OPEN_ACCESSBOTH); *tl = txdr_unsigned((mode >> NFSLCK_SHIFT) & NFSV4OPEN_DENYBOTH); error = nfscl_request(nd, vp, p, cred, NULL); if (error) return (error); NFSCL_INCRSEQID(op->nfso_own->nfsow_seqid, nd); if (!nd->nd_repstat) { NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID); op->nfso_stateid.seqid = *tl++; op->nfso_stateid.other[0] = *tl++; op->nfso_stateid.other[1] = *tl++; op->nfso_stateid.other[2] = *tl; } if (nd->nd_repstat && error == 0) error = nd->nd_repstat; if (error == NFSERR_STALESTATEID) nfscl_initiate_recovery(op->nfso_own->nfsow_clp); nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * V4 Close operation. */ APPLESTATIC int nfsrpc_close(vnode_t vp, int doclose, NFSPROC_T *p) { struct nfsclclient *clp; int error; if (vnode_vtype(vp) != VREG) return (0); if (doclose) error = nfscl_doclose(vp, &clp, p); else error = nfscl_getclose(vp, &clp); if (error) return (error); nfscl_clientrelease(clp); return (0); } /* * Close the open. */ APPLESTATIC void nfsrpc_doclose(struct nfsmount *nmp, struct nfsclopen *op, NFSPROC_T *p) { struct nfsrv_descript nfsd, *nd = &nfsd; struct nfscllockowner *lp, *nlp; struct nfscllock *lop, *nlop; struct ucred *tcred; u_int64_t off = 0, len = 0; u_int32_t type = NFSV4LOCKT_READ; int error, do_unlock, trycnt; tcred = newnfs_getcred(); newnfs_copycred(&op->nfso_cred, tcred); /* * (Theoretically this could be done in the same * compound as the close, but having multiple * sequenced Ops in the same compound might be * too scary for some servers.) */ if (op->nfso_posixlock) { off = 0; len = NFS64BITSSET; type = NFSV4LOCKT_READ; } /* * Since this function is only called from VOP_INACTIVE(), no * other thread will be manipulating this Open. As such, the * lock lists are not being changed by other threads, so it should * be safe to do this without locking. */ LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) { do_unlock = 1; LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) { if (op->nfso_posixlock == 0) { off = lop->nfslo_first; len = lop->nfslo_end - lop->nfslo_first; if (lop->nfslo_type == F_WRLCK) type = NFSV4LOCKT_WRITE; else type = NFSV4LOCKT_READ; } if (do_unlock) { trycnt = 0; do { error = nfsrpc_locku(nd, nmp, lp, off, len, type, tcred, p, 0); if ((nd->nd_repstat == NFSERR_GRACE || nd->nd_repstat == NFSERR_DELAY) && error == 0) (void) nfs_catnap(PZERO, (int)nd->nd_repstat, "nfs_close"); } while ((nd->nd_repstat == NFSERR_GRACE || nd->nd_repstat == NFSERR_DELAY) && error == 0 && trycnt++ < 5); if (op->nfso_posixlock) do_unlock = 0; } nfscl_freelock(lop, 0); } /* * Do a ReleaseLockOwner. * The lock owner name nfsl_owner may be used by other opens for * other files but the lock_owner4 name that nfsrpc_rellockown() * puts on the wire has the file handle for this file appended * to it, so it can be done now. */ (void)nfsrpc_rellockown(nmp, lp, lp->nfsl_open->nfso_fh, lp->nfsl_open->nfso_fhlen, tcred, p); } /* * There could be other Opens for different files on the same * OpenOwner, so locking is required. */ NFSLOCKCLSTATE(); nfscl_lockexcl(&op->nfso_own->nfsow_rwlock, NFSCLSTATEMUTEXPTR); NFSUNLOCKCLSTATE(); do { error = nfscl_tryclose(op, tcred, nmp, p); if (error == NFSERR_GRACE) (void) nfs_catnap(PZERO, error, "nfs_close"); } while (error == NFSERR_GRACE); NFSLOCKCLSTATE(); nfscl_lockunlock(&op->nfso_own->nfsow_rwlock); LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) nfscl_freelockowner(lp, 0); nfscl_freeopen(op, 0); NFSUNLOCKCLSTATE(); NFSFREECRED(tcred); } /* * The actual Close RPC. */ APPLESTATIC int nfsrpc_closerpc(struct nfsrv_descript *nd, struct nfsmount *nmp, struct nfsclopen *op, struct ucred *cred, NFSPROC_T *p, int syscred) { u_int32_t *tl; int error; nfscl_reqstart(nd, NFSPROC_CLOSE, nmp, op->nfso_fh, op->nfso_fhlen, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID); *tl++ = txdr_unsigned(op->nfso_own->nfsow_seqid); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = op->nfso_stateid.seqid; *tl++ = op->nfso_stateid.other[0]; *tl++ = op->nfso_stateid.other[1]; *tl = op->nfso_stateid.other[2]; if (syscred) nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); NFSCL_INCRSEQID(op->nfso_own->nfsow_seqid, nd); if (nd->nd_repstat == 0) NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID); error = nd->nd_repstat; if (error == NFSERR_STALESTATEID) nfscl_initiate_recovery(op->nfso_own->nfsow_clp); nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * V4 Open Confirm RPC. */ APPLESTATIC int nfsrpc_openconfirm(vnode_t vp, u_int8_t *nfhp, int fhlen, struct nfsclopen *op, struct ucred *cred, NFSPROC_T *p) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp; int error; nmp = VFSTONFS(vnode_mount(vp)); if (NFSHASNFSV4N(nmp)) return (0); /* No confirmation for NFSv4.1. */ nfscl_reqstart(nd, NFSPROC_OPENCONFIRM, nmp, nfhp, fhlen, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID); *tl++ = op->nfso_stateid.seqid; *tl++ = op->nfso_stateid.other[0]; *tl++ = op->nfso_stateid.other[1]; *tl++ = op->nfso_stateid.other[2]; *tl = txdr_unsigned(op->nfso_own->nfsow_seqid); error = nfscl_request(nd, vp, p, cred, NULL); if (error) return (error); NFSCL_INCRSEQID(op->nfso_own->nfsow_seqid, nd); if (!nd->nd_repstat) { NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID); op->nfso_stateid.seqid = *tl++; op->nfso_stateid.other[0] = *tl++; op->nfso_stateid.other[1] = *tl++; op->nfso_stateid.other[2] = *tl; } error = nd->nd_repstat; if (error == NFSERR_STALESTATEID) nfscl_initiate_recovery(op->nfso_own->nfsow_clp); nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Do the setclientid and setclientid confirm RPCs. Called from nfs_statfs() * when a mount has just occurred and when the server replies NFSERR_EXPIRED. */ APPLESTATIC int nfsrpc_setclient(struct nfsmount *nmp, struct nfsclclient *clp, int reclaim, struct ucred *cred, NFSPROC_T *p) { u_int32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; nfsattrbit_t attrbits; u_int8_t *cp = NULL, *cp2, addr[INET6_ADDRSTRLEN + 9]; u_short port; int error, isinet6 = 0, callblen; nfsquad_t confirm; u_int32_t lease; static u_int32_t rev = 0; struct nfsclds *dsp; struct in6_addr a6; struct nfsclsession *tsep; if (nfsboottime.tv_sec == 0) NFSSETBOOTTIME(nfsboottime); clp->nfsc_rev = rev++; if (NFSHASNFSV4N(nmp)) { /* * Either there was no previous session or the * previous session has failed, so... * do an ExchangeID followed by the CreateSession. */ error = nfsrpc_exchangeid(nmp, clp, &nmp->nm_sockreq, NFSV4EXCH_USEPNFSMDS | NFSV4EXCH_USENONPNFS, &dsp, cred, p); NFSCL_DEBUG(1, "aft exch=%d\n", error); if (error == 0) error = nfsrpc_createsession(nmp, &dsp->nfsclds_sess, &nmp->nm_sockreq, dsp->nfsclds_sess.nfsess_sequenceid, 1, cred, p); if (error == 0) { NFSLOCKMNT(nmp); /* * The old sessions cannot be safely free'd * here, since they may still be used by * in-progress RPCs. */ tsep = NULL; if (TAILQ_FIRST(&nmp->nm_sess) != NULL) tsep = NFSMNT_MDSSESSION(nmp); TAILQ_INSERT_HEAD(&nmp->nm_sess, dsp, nfsclds_list); /* * Wake up RPCs waiting for a slot on the * old session. These will then fail with * NFSERR_BADSESSION and be retried with the * new session by nfsv4_setsequence(). * Also wakeup() processes waiting for the * new session. */ if (tsep != NULL) wakeup(&tsep->nfsess_slots); wakeup(&nmp->nm_sess); NFSUNLOCKMNT(nmp); } else nfscl_freenfsclds(dsp); NFSCL_DEBUG(1, "aft createsess=%d\n", error); if (error == 0 && reclaim == 0) { error = nfsrpc_reclaimcomplete(nmp, cred, p); NFSCL_DEBUG(1, "aft reclaimcomp=%d\n", error); if (error == NFSERR_COMPLETEALREADY || error == NFSERR_NOTSUPP) /* Ignore this error. */ error = 0; } return (error); } /* * Allocate a single session structure for NFSv4.0, because some of * the fields are used by NFSv4.0 although it doesn't do a session. */ dsp = malloc(sizeof(struct nfsclds), M_NFSCLDS, M_WAITOK | M_ZERO); mtx_init(&dsp->nfsclds_mtx, "nfsds", NULL, MTX_DEF); mtx_init(&dsp->nfsclds_sess.nfsess_mtx, "nfssession", NULL, MTX_DEF); NFSLOCKMNT(nmp); TAILQ_INSERT_HEAD(&nmp->nm_sess, dsp, nfsclds_list); tsep = NFSMNT_MDSSESSION(nmp); NFSUNLOCKMNT(nmp); nfscl_reqstart(nd, NFSPROC_SETCLIENTID, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(nfsboottime.tv_sec); *tl = txdr_unsigned(clp->nfsc_rev); (void) nfsm_strtom(nd, clp->nfsc_id, clp->nfsc_idlen); /* * set up the callback address */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFS_CALLBCKPROG); callblen = strlen(nfsv4_callbackaddr); if (callblen == 0) cp = nfscl_getmyip(nmp, &a6, &isinet6); if (nfscl_enablecallb && nfs_numnfscbd > 0 && (callblen > 0 || cp != NULL)) { port = htons(nfsv4_cbport); cp2 = (u_int8_t *)&port; #ifdef INET6 if ((callblen > 0 && strchr(nfsv4_callbackaddr, ':')) || isinet6) { char ip6buf[INET6_ADDRSTRLEN], *ip6add; (void) nfsm_strtom(nd, "tcp6", 4); if (callblen == 0) { ip6_sprintf(ip6buf, (struct in6_addr *)cp); ip6add = ip6buf; } else { ip6add = nfsv4_callbackaddr; } snprintf(addr, INET6_ADDRSTRLEN + 9, "%s.%d.%d", ip6add, cp2[0], cp2[1]); } else #endif { (void) nfsm_strtom(nd, "tcp", 3); if (callblen == 0) snprintf(addr, INET6_ADDRSTRLEN + 9, "%d.%d.%d.%d.%d.%d", cp[0], cp[1], cp[2], cp[3], cp2[0], cp2[1]); else snprintf(addr, INET6_ADDRSTRLEN + 9, "%s.%d.%d", nfsv4_callbackaddr, cp2[0], cp2[1]); } (void) nfsm_strtom(nd, addr, strlen(addr)); } else { (void) nfsm_strtom(nd, "tcp", 3); (void) nfsm_strtom(nd, "0.0.0.0.0.0", 11); } NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(clp->nfsc_cbident); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED); tsep->nfsess_clientid.lval[0] = *tl++; tsep->nfsess_clientid.lval[1] = *tl++; confirm.lval[0] = *tl++; confirm.lval[1] = *tl; mbuf_freem(nd->nd_mrep); nd->nd_mrep = NULL; /* * and confirm it. */ nfscl_reqstart(nd, NFSPROC_SETCLIENTIDCFRM, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); *tl++ = tsep->nfsess_clientid.lval[0]; *tl++ = tsep->nfsess_clientid.lval[1]; *tl++ = confirm.lval[0]; *tl = confirm.lval[1]; nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); mbuf_freem(nd->nd_mrep); nd->nd_mrep = NULL; if (nd->nd_repstat == 0) { nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, nmp->nm_fh, nmp->nm_fhsize, NULL, NULL); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_LEASETIME); (void) nfsrv_putattrbit(nd, &attrbits); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); if (nd->nd_repstat == 0) { error = nfsv4_loadattr(nd, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, &lease, NULL, p, cred); if (error) goto nfsmout; 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++; } } } error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs getattr call. */ APPLESTATIC int nfsrpc_getattr(vnode_t vp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, void *stuff) { struct nfsrv_descript nfsd, *nd = &nfsd; int error; nfsattrbit_t attrbits; NFSCL_REQSTART(nd, NFSPROC_GETATTR, vp); if (nd->nd_flag & ND_NFSV4) { NFSGETATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (!nd->nd_repstat) error = nfsm_loadattr(nd, nap); else error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * nfs getattr call with non-vnode arguemnts. */ APPLESTATIC int nfsrpc_getattrnovp(struct nfsmount *nmp, u_int8_t *fhp, int fhlen, int syscred, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, u_int64_t *xidp, uint32_t *leasep) { struct nfsrv_descript nfsd, *nd = &nfsd; int error, vers = NFS_VER2; nfsattrbit_t attrbits; nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, fhp, fhlen, NULL, NULL); if (nd->nd_flag & ND_NFSV4) { vers = NFS_VER4; NFSGETATTR_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_LEASETIME); (void) nfsrv_putattrbit(nd, &attrbits); } else if (nd->nd_flag & ND_NFSV3) { vers = NFS_VER3; } if (syscred) nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, vers, NULL, 1, xidp, NULL); if (error) return (error); if (nd->nd_repstat == 0) { if ((nd->nd_flag & ND_NFSV4) != 0) error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, leasep, NULL, NULL, NULL); else error = nfsm_loadattr(nd, nap); } else error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * Do an nfs setattr operation. */ APPLESTATIC int nfsrpc_setattr(vnode_t vp, struct vattr *vap, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *rnap, int *attrflagp, void *stuff) { int error, expireret = 0, openerr, retrycnt; u_int32_t clidrev = 0, mode; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsfh *nfhp; nfsv4stateid_t stateid; void *lckp; if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; if (vap != NULL && NFSATTRISSET(u_quad_t, vap, va_size)) mode = NFSV4OPEN_ACCESSWRITE; else mode = NFSV4OPEN_ACCESSREAD; retrycnt = 0; do { lckp = NULL; openerr = 1; if (NFSHASNFSV4(nmp)) { nfhp = VTONFS(vp)->n_fhp; error = nfscl_getstateid(vp, nfhp->nfh_fh, nfhp->nfh_len, mode, 0, cred, p, &stateid, &lckp); if (error && vnode_vtype(vp) == VREG && (mode == NFSV4OPEN_ACCESSWRITE || nfstest_openallsetattr)) { /* * No Open stateid, so try and open the file * now. */ if (mode == NFSV4OPEN_ACCESSWRITE) openerr = nfsrpc_open(vp, FWRITE, cred, p); else openerr = nfsrpc_open(vp, FREAD, cred, p); if (!openerr) (void) nfscl_getstateid(vp, nfhp->nfh_fh, nfhp->nfh_len, mode, 0, cred, p, &stateid, &lckp); } } if (vap != NULL) error = nfsrpc_setattrrpc(vp, vap, &stateid, cred, p, rnap, attrflagp, stuff); else error = nfsrpc_setaclrpc(vp, cred, p, aclp, &stateid, stuff); if (error == NFSERR_OPENMODE && mode == NFSV4OPEN_ACCESSREAD) { NFSLOCKMNT(nmp); nmp->nm_state |= NFSSTA_OPENMODE; NFSUNLOCKMNT(nmp); } if (error == NFSERR_STALESTATEID) nfscl_initiate_recovery(nmp->nm_clp); if (lckp != NULL) nfscl_lockderef(lckp); if (!openerr) (void) nfsrpc_close(vp, 0, p); if (error == NFSERR_GRACE || error == NFSERR_STALESTATEID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_OLDSTATEID || error == NFSERR_BADSESSION) { (void) nfs_catnap(PZERO, error, "nfs_setattr"); } else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && clidrev != 0) { expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p); } retrycnt++; } while (error == NFSERR_GRACE || error == NFSERR_STALESTATEID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_BADSESSION || (error == NFSERR_OLDSTATEID && retrycnt < 20) || ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && expireret == 0 && clidrev != 0 && retrycnt < 4) || (error == NFSERR_OPENMODE && mode == NFSV4OPEN_ACCESSREAD && retrycnt < 4)); if (error && retrycnt >= 4) error = EIO; return (error); } static int nfsrpc_setattrrpc(vnode_t vp, struct vattr *vap, nfsv4stateid_t *stateidp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *rnap, int *attrflagp, void *stuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; int error; nfsattrbit_t attrbits; *attrflagp = 0; NFSCL_REQSTART(nd, NFSPROC_SETATTR, vp); if (nd->nd_flag & ND_NFSV4) nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); vap->va_type = vnode_vtype(vp); nfscl_fillsattr(nd, vap, vp, NFSSATTR_FULL, 0); if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = newnfs_false; } else if (nd->nd_flag & ND_NFSV4) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) error = nfscl_wcc_data(nd, vp, rnap, attrflagp, NULL, stuff); if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4 && !error) error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (!(nd->nd_flag & ND_NFSV3) && !nd->nd_repstat && !error) error = nfscl_postop_attr(nd, rnap, attrflagp, stuff); mbuf_freem(nd->nd_mrep); if (nd->nd_repstat && !error) error = nd->nd_repstat; return (error); } /* * nfs lookup rpc */ APPLESTATIC int nfsrpc_lookup(vnode_t dvp, char *name, int len, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *stuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp; struct nfsnode *np; struct nfsfh *nfhp; nfsattrbit_t attrbits; int error = 0, lookupp = 0; *attrflagp = 0; *dattrflagp = 0; if (vnode_vtype(dvp) != VDIR) return (ENOTDIR); nmp = VFSTONFS(vnode_mount(dvp)); if (len > NFS_MAXNAMLEN) return (ENAMETOOLONG); if (NFSHASNFSV4(nmp) && len == 1 && name[0] == '.') { /* * Just return the current dir's fh. */ np = VTONFS(dvp); MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + np->n_fhp->nfh_len, M_NFSFH, M_WAITOK); nfhp->nfh_len = np->n_fhp->nfh_len; NFSBCOPY(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len); *nfhpp = nfhp; return (0); } if (NFSHASNFSV4(nmp) && len == 2 && name[0] == '.' && name[1] == '.') { lookupp = 1; NFSCL_REQSTART(nd, NFSPROC_LOOKUPP, dvp); } else { NFSCL_REQSTART(nd, NFSPROC_LOOKUP, dvp); (void) nfsm_strtom(nd, name, len); } if (nd->nd_flag & ND_NFSV4) { NFSGETATTR_ATTRBIT(&attrbits); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OP_GETFH); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, dvp, p, cred, stuff); if (error) return (error); if (nd->nd_repstat) { /* * When an NFSv4 Lookupp returns ENOENT, it means that * the lookup is at the root of an fs, so return this dir. */ if (nd->nd_repstat == NFSERR_NOENT && lookupp) { np = VTONFS(dvp); MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + np->n_fhp->nfh_len, M_NFSFH, M_WAITOK); nfhp->nfh_len = np->n_fhp->nfh_len; NFSBCOPY(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len); *nfhpp = nfhp; mbuf_freem(nd->nd_mrep); return (0); } if (nd->nd_flag & ND_NFSV3) error = nfscl_postop_attr(nd, dnap, dattrflagp, stuff); else if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4) { /* Load the directory attributes. */ error = nfsm_loadattr(nd, dnap); if (error == 0) *dattrflagp = 1; } goto nfsmout; } if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4) { /* Load the directory attributes. */ error = nfsm_loadattr(nd, dnap); if (error != 0) goto nfsmout; *dattrflagp = 1; /* Skip over the Lookup and GetFH operation status values. */ NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED); } error = nfsm_getfh(nd, nfhpp); if (error) goto nfsmout; error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if ((nd->nd_flag & ND_NFSV3) && !error) error = nfscl_postop_attr(nd, dnap, dattrflagp, stuff); nfsmout: mbuf_freem(nd->nd_mrep); if (!error && nd->nd_repstat) error = nd->nd_repstat; return (error); } /* * Do a readlink rpc. */ APPLESTATIC int nfsrpc_readlink(vnode_t vp, struct uio *uiop, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsnode *np = VTONFS(vp); nfsattrbit_t attrbits; int error, len, cangetattr = 1; *attrflagp = 0; NFSCL_REQSTART(nd, NFSPROC_READLINK, vp); if (nd->nd_flag & ND_NFSV4) { /* * And do a Getattr op. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_flag & ND_NFSV3) error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (!nd->nd_repstat && !error) { NFSM_STRSIZ(len, NFS_MAXPATHLEN); /* * This seems weird to me, but must have been added to * FreeBSD for some reason. The only thing I can think of * is that there was/is some server that replies with * more link data than it should? */ if (len == NFS_MAXPATHLEN) { NFSLOCKNODE(np); if (np->n_size > 0 && np->n_size < NFS_MAXPATHLEN) { len = np->n_size; cangetattr = 0; } NFSUNLOCKNODE(np); } error = nfsm_mbufuio(nd, uiop, len); if ((nd->nd_flag & ND_NFSV4) && !error && cangetattr) error = nfscl_postop_attr(nd, nap, attrflagp, stuff); } if (nd->nd_repstat && !error) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Read operation. */ APPLESTATIC int nfsrpc_read(vnode_t vp, struct uio *uiop, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { int error, expireret = 0, retrycnt; u_int32_t clidrev = 0; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsnode *np = VTONFS(vp); struct ucred *newcred; struct nfsfh *nfhp = NULL; nfsv4stateid_t stateid; void *lckp; if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; newcred = cred; if (NFSHASNFSV4(nmp)) { nfhp = np->n_fhp; newcred = NFSNEWCRED(cred); } retrycnt = 0; do { lckp = NULL; if (NFSHASNFSV4(nmp)) (void)nfscl_getstateid(vp, nfhp->nfh_fh, nfhp->nfh_len, NFSV4OPEN_ACCESSREAD, 0, newcred, p, &stateid, &lckp); error = nfsrpc_readrpc(vp, uiop, newcred, &stateid, p, nap, attrflagp, stuff); if (error == NFSERR_OPENMODE) { NFSLOCKMNT(nmp); nmp->nm_state |= NFSSTA_OPENMODE; NFSUNLOCKMNT(nmp); } if (error == NFSERR_STALESTATEID) nfscl_initiate_recovery(nmp->nm_clp); if (lckp != NULL) nfscl_lockderef(lckp); if (error == NFSERR_GRACE || error == NFSERR_STALESTATEID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_OLDSTATEID || error == NFSERR_BADSESSION) { (void) nfs_catnap(PZERO, error, "nfs_read"); } else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && clidrev != 0) { expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p); } retrycnt++; } while (error == NFSERR_GRACE || error == NFSERR_STALESTATEID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_BADSESSION || (error == NFSERR_OLDSTATEID && retrycnt < 20) || ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && expireret == 0 && clidrev != 0 && retrycnt < 4) || (error == NFSERR_OPENMODE && retrycnt < 4)); if (error && retrycnt >= 4) error = EIO; if (NFSHASNFSV4(nmp)) NFSFREECRED(newcred); return (error); } /* * The actual read RPC. */ static int nfsrpc_readrpc(vnode_t vp, struct uio *uiop, struct ucred *cred, nfsv4stateid_t *stateidp, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { u_int32_t *tl; int error = 0, len, retlen, tsiz, eof = 0; struct nfsrv_descript nfsd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsrv_descript *nd = &nfsd; int rsize; off_t tmp_off; *attrflagp = 0; tsiz = uio_uio_resid(uiop); tmp_off = uiop->uio_offset + tsiz; NFSLOCKMNT(nmp); if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) { NFSUNLOCKMNT(nmp); return (EFBIG); } rsize = nmp->nm_rsize; NFSUNLOCKMNT(nmp); nd->nd_mrep = NULL; while (tsiz > 0) { *attrflagp = 0; len = (tsiz > rsize) ? rsize : tsiz; NFSCL_REQSTART(nd, NFSPROC_READ, vp); if (nd->nd_flag & ND_NFSV4) nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED * 3); if (nd->nd_flag & ND_NFSV2) { *tl++ = txdr_unsigned(uiop->uio_offset); *tl++ = txdr_unsigned(len); *tl = 0; } else { txdr_hyper(uiop->uio_offset, tl); *(tl + 2) = txdr_unsigned(len); } /* * Since I can't do a Getattr for NFSv4 for Write, there * doesn't seem any point in doing one here, either. * (See the comment in nfsrpc_writerpc() for more info.) */ error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_flag & ND_NFSV3) { error = nfscl_postop_attr(nd, nap, attrflagp, stuff); } else if (!nd->nd_repstat && (nd->nd_flag & ND_NFSV2)) { error = nfsm_loadattr(nd, nap); if (!error) *attrflagp = 1; } if (nd->nd_repstat || error) { if (!error) error = nd->nd_repstat; goto nfsmout; } if (nd->nd_flag & ND_NFSV3) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); eof = fxdr_unsigned(int, *(tl + 1)); } else if (nd->nd_flag & ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); eof = fxdr_unsigned(int, *tl); } NFSM_STRSIZ(retlen, len); error = nfsm_mbufuio(nd, uiop, retlen); if (error) goto nfsmout; mbuf_freem(nd->nd_mrep); nd->nd_mrep = NULL; tsiz -= retlen; if (!(nd->nd_flag & ND_NFSV2)) { if (eof || retlen == 0) tsiz = 0; } else if (retlen < len) tsiz = 0; } return (0); nfsmout: if (nd->nd_mrep != NULL) mbuf_freem(nd->nd_mrep); return (error); } /* * nfs write operation * When called_from_strategy != 0, it should return EIO for an error that * indicates recovery is in progress, so that the buffer will be left * dirty and be written back to the server later. If it loops around, * the recovery thread could get stuck waiting for the buffer and recovery * will then deadlock. */ APPLESTATIC int nfsrpc_write(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff, int called_from_strategy) { int error, expireret = 0, retrycnt, nostateid; u_int32_t clidrev = 0; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsnode *np = VTONFS(vp); struct ucred *newcred; struct nfsfh *nfhp = NULL; nfsv4stateid_t stateid; void *lckp; *must_commit = 0; if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; newcred = cred; if (NFSHASNFSV4(nmp)) { newcred = NFSNEWCRED(cred); nfhp = np->n_fhp; } retrycnt = 0; do { lckp = NULL; nostateid = 0; if (NFSHASNFSV4(nmp)) { (void)nfscl_getstateid(vp, nfhp->nfh_fh, nfhp->nfh_len, NFSV4OPEN_ACCESSWRITE, 0, newcred, p, &stateid, &lckp); if (stateid.other[0] == 0 && stateid.other[1] == 0 && stateid.other[2] == 0) { nostateid = 1; NFSCL_DEBUG(1, "stateid0 in write\n"); } } /* * If there is no stateid for NFSv4, it means this is an * extraneous write after close. Basically a poorly * implemented buffer cache. Just don't do the write. */ if (nostateid) error = 0; else error = nfsrpc_writerpc(vp, uiop, iomode, must_commit, newcred, &stateid, p, nap, attrflagp, stuff); if (error == NFSERR_STALESTATEID) nfscl_initiate_recovery(nmp->nm_clp); if (lckp != NULL) nfscl_lockderef(lckp); if (error == NFSERR_GRACE || error == NFSERR_STALESTATEID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_OLDSTATEID || error == NFSERR_BADSESSION) { (void) nfs_catnap(PZERO, error, "nfs_write"); } else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && clidrev != 0) { expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p); } retrycnt++; } while (error == NFSERR_GRACE || error == NFSERR_DELAY || ((error == NFSERR_STALESTATEID || error == NFSERR_BADSESSION || error == NFSERR_STALEDONTRECOVER) && called_from_strategy == 0) || (error == NFSERR_OLDSTATEID && retrycnt < 20) || ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && expireret == 0 && clidrev != 0 && retrycnt < 4)); if (error != 0 && (retrycnt >= 4 || ((error == NFSERR_STALESTATEID || error == NFSERR_BADSESSION || error == NFSERR_STALEDONTRECOVER) && called_from_strategy != 0))) error = EIO; if (NFSHASNFSV4(nmp)) NFSFREECRED(newcred); return (error); } /* * The actual write RPC. */ static int nfsrpc_writerpc(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, struct ucred *cred, nfsv4stateid_t *stateidp, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { u_int32_t *tl; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsnode *np = VTONFS(vp); int error = 0, len, tsiz, rlen, commit, committed = NFSWRITE_FILESYNC; int wccflag = 0, wsize; int32_t backup; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; nfsattrbit_t attrbits; off_t tmp_off; KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1")); *attrflagp = 0; tsiz = uio_uio_resid(uiop); tmp_off = uiop->uio_offset + tsiz; NFSLOCKMNT(nmp); if (tmp_off > nmp->nm_maxfilesize || tmp_off < uiop->uio_offset) { NFSUNLOCKMNT(nmp); return (EFBIG); } wsize = nmp->nm_wsize; NFSUNLOCKMNT(nmp); nd->nd_mrep = NULL; /* NFSv2 sometimes does a write with */ nd->nd_repstat = 0; /* uio_resid == 0, so the while is not done */ while (tsiz > 0) { *attrflagp = 0; len = (tsiz > wsize) ? wsize : tsiz; NFSCL_REQSTART(nd, NFSPROC_WRITE, vp); if (nd->nd_flag & ND_NFSV4) { nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER+2*NFSX_UNSIGNED); txdr_hyper(uiop->uio_offset, tl); tl += 2; *tl++ = txdr_unsigned(*iomode); *tl = txdr_unsigned(len); } else if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER+3*NFSX_UNSIGNED); txdr_hyper(uiop->uio_offset, tl); tl += 2; *tl++ = txdr_unsigned(len); *tl++ = txdr_unsigned(*iomode); *tl = txdr_unsigned(len); } else { u_int32_t x; NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); /* * Not sure why someone changed this, since the * RFC clearly states that "beginoffset" and * "totalcount" are ignored, but it wouldn't * surprise me if there's a busted server out there. */ /* Set both "begin" and "current" to non-garbage. */ x = txdr_unsigned((u_int32_t)uiop->uio_offset); *tl++ = x; /* "begin offset" */ *tl++ = x; /* "current offset" */ x = txdr_unsigned(len); *tl++ = x; /* total to this offset */ *tl = x; /* size of this write */ } nfsm_uiombuf(nd, uiop, len); /* * Although it is tempting to do a normal Getattr Op in the * NFSv4 compound, the result can be a nearly hung client * system if the Getattr asks for Owner and/or OwnerGroup. * It occurs when the client can't map either the Owner or * Owner_group name in the Getattr reply to a uid/gid. When * there is a cache miss, the kernel does an upcall to the * nfsuserd. Then, it can try and read the local /etc/passwd * or /etc/group file. It can then block in getnewbuf(), * waiting for dirty writes to be pushed to the NFS server. * The only reason this doesn't result in a complete * deadlock, is that the upcall times out and allows * the write to complete. However, progress is so slow * that it might just as well be deadlocked. * As such, we get the rest of the attributes, but not * Owner or Owner_group. * nb: nfscl_loadattrcache() needs to be told that these * partial attributes from a write rpc are being * passed in, via a argument flag. */ if (nd->nd_flag & ND_NFSV4) { NFSWRITEGETATTR_ATTRBIT(&attrbits); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_repstat) { /* * In case the rpc gets retried, roll * the uio fileds changed by nfsm_uiombuf() * back. */ uiop->uio_offset -= len; uio_uio_resid_add(uiop, len); uio_iov_base_add(uiop, -len); uio_iov_len_add(uiop, len); } if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { error = nfscl_wcc_data(nd, vp, nap, attrflagp, &wccflag, stuff); if (error) goto nfsmout; } if (!nd->nd_repstat) { if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); rlen = fxdr_unsigned(int, *tl++); if (rlen == 0) { error = NFSERR_IO; goto nfsmout; } else if (rlen < len) { backup = len - rlen; uio_iov_base_add(uiop, -(backup)); uio_iov_len_add(uiop, backup); uiop->uio_offset -= backup; uio_uio_resid_add(uiop, backup); len = rlen; } commit = fxdr_unsigned(int, *tl++); /* * Return the lowest commitment level * obtained by any of the RPCs. */ if (committed == NFSWRITE_FILESYNC) committed = commit; else if (committed == NFSWRITE_DATASYNC && commit == NFSWRITE_UNSTABLE) committed = commit; NFSLOCKMNT(nmp); if (!NFSHASWRITEVERF(nmp)) { NFSBCOPY((caddr_t)tl, (caddr_t)&nmp->nm_verf[0], NFSX_VERF); NFSSETWRITEVERF(nmp); } else if (NFSBCMP(tl, nmp->nm_verf, NFSX_VERF)) { *must_commit = 1; NFSBCOPY(tl, nmp->nm_verf, NFSX_VERF); } NFSUNLOCKMNT(nmp); } if (nd->nd_flag & ND_NFSV4) NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (nd->nd_flag & (ND_NFSV2 | ND_NFSV4)) { error = nfsm_loadattr(nd, nap); if (!error) *attrflagp = NFS_LATTR_NOSHRINK; } } else { error = nd->nd_repstat; } if (error) goto nfsmout; NFSWRITERPC_SETTIME(wccflag, np, nap, (nd->nd_flag & ND_NFSV4)); mbuf_freem(nd->nd_mrep); nd->nd_mrep = NULL; tsiz -= len; } nfsmout: if (nd->nd_mrep != NULL) mbuf_freem(nd->nd_mrep); *iomode = committed; if (nd->nd_repstat && !error) error = nd->nd_repstat; 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. */ APPLESTATIC int nfsrpc_mknod(vnode_t dvp, char *name, int namelen, struct vattr *vap, u_int32_t rdev, enum vtype vtyp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff) { u_int32_t *tl; int error = 0; struct nfsrv_descript nfsd, *nd = &nfsd; nfsattrbit_t attrbits; *nfhpp = NULL; *attrflagp = 0; *dattrflagp = 0; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_MKNOD, dvp); if (nd->nd_flag & ND_NFSV4) { if (vtyp == VBLK || vtyp == VCHR) { NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); *tl++ = vtonfsv34_type(vtyp); *tl++ = txdr_unsigned(NFSMAJOR(rdev)); *tl = txdr_unsigned(NFSMINOR(rdev)); } else { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = vtonfsv34_type(vtyp); } } (void) nfsm_strtom(nd, name, namelen); if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = vtonfsv34_type(vtyp); } if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) nfscl_fillsattr(nd, vap, dvp, 0, 0); if ((nd->nd_flag & ND_NFSV3) && (vtyp == VCHR || vtyp == VBLK)) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSMAJOR(rdev)); *tl = txdr_unsigned(NFSMINOR(rdev)); } if (nd->nd_flag & ND_NFSV4) { NFSGETATTR_ATTRBIT(&attrbits); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OP_GETFH); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); } if (nd->nd_flag & ND_NFSV2) nfscl_fillsattr(nd, vap, dvp, NFSSATTR_SIZERDEV, rdev); error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); if (nd->nd_flag & ND_NFSV4) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if (!nd->nd_repstat) { if (nd->nd_flag & ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); if (error) goto nfsmout; } error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp); if (error) goto nfsmout; } if (nd->nd_flag & ND_NFSV3) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if (!error && nd->nd_repstat) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs file create call * Mostly just call the approriate routine. (I separated out v4, so that * error recovery wouldn't be as difficult.) */ APPLESTATIC int nfsrpc_create(vnode_t dvp, char *name, int namelen, struct vattr *vap, nfsquad_t cverf, int fmode, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff) { int error = 0, newone, expireret = 0, retrycnt, unlocked; struct nfsclowner *owp; struct nfscldeleg *dp; struct nfsmount *nmp = VFSTONFS(vnode_mount(dvp)); u_int32_t clidrev; if (NFSHASNFSV4(nmp)) { retrycnt = 0; do { dp = NULL; error = nfscl_open(dvp, NULL, 0, (NFSV4OPEN_ACCESSWRITE | NFSV4OPEN_ACCESSREAD), 0, cred, p, &owp, NULL, &newone, NULL, 1); if (error) return (error); if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; else clidrev = 0; error = nfsrpc_createv4(dvp, name, namelen, vap, cverf, fmode, owp, &dp, cred, p, dnap, nnap, nfhpp, attrflagp, dattrflagp, dstuff, &unlocked); /* * There is no need to invalidate cached attributes here, * since new post-delegation issue attributes are always * returned by nfsrpc_createv4() and these will update the * attribute cache. */ if (dp != NULL) (void) nfscl_deleg(nmp->nm_mountp, owp->nfsow_clp, (*nfhpp)->nfh_fh, (*nfhpp)->nfh_len, cred, p, &dp); nfscl_ownerrelease(nmp, owp, error, newone, unlocked); if (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_BADSESSION) { (void) nfs_catnap(PZERO, error, "nfs_open"); } else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && clidrev != 0) { expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p); retrycnt++; } } while (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_DELAY || error == NFSERR_BADSESSION || ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && expireret == 0 && clidrev != 0 && retrycnt < 4)); if (error && retrycnt >= 4) error = EIO; } else { error = nfsrpc_createv23(dvp, name, namelen, vap, cverf, fmode, cred, p, dnap, nnap, nfhpp, attrflagp, dattrflagp, dstuff); } return (error); } /* * The create rpc for v2 and 3. */ static int nfsrpc_createv23(vnode_t dvp, char *name, int namelen, struct vattr *vap, nfsquad_t cverf, int fmode, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff) { u_int32_t *tl; int error = 0; struct nfsrv_descript nfsd, *nd = &nfsd; *nfhpp = NULL; *attrflagp = 0; *dattrflagp = 0; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_CREATE, dvp); (void) nfsm_strtom(nd, name, namelen); if (nd->nd_flag & ND_NFSV3) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); if (fmode & O_EXCL) { *tl = txdr_unsigned(NFSCREATE_EXCLUSIVE); NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); *tl++ = cverf.lval[0]; *tl = cverf.lval[1]; } else { *tl = txdr_unsigned(NFSCREATE_UNCHECKED); nfscl_fillsattr(nd, vap, dvp, 0, 0); } } else { nfscl_fillsattr(nd, vap, dvp, NFSSATTR_SIZE0, 0); } error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); if (nd->nd_repstat == 0) { error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp); if (error) goto nfsmout; } if (nd->nd_flag & ND_NFSV3) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if (nd->nd_repstat != 0 && error == 0) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } static int nfsrpc_createv4(vnode_t dvp, char *name, int namelen, struct vattr *vap, nfsquad_t cverf, int fmode, struct nfsclowner *owp, struct nfscldeleg **dpp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff, int *unlockedp) { u_int32_t *tl; int error = 0, deleg, newone, ret, acesize, limitby; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsclopen *op; struct nfscldeleg *dp = NULL; struct nfsnode *np; struct nfsfh *nfhp; nfsattrbit_t attrbits; nfsv4stateid_t stateid; u_int32_t rflags; struct nfsmount *nmp; struct nfsclsession *tsep; nmp = VFSTONFS(dvp->v_mount); np = VTONFS(dvp); *unlockedp = 0; *nfhpp = NULL; *dpp = NULL; *attrflagp = 0; *dattrflagp = 0; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_CREATE, dvp); /* * For V4, this is actually an Open op. */ NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(owp->nfsow_seqid); *tl++ = txdr_unsigned(NFSV4OPEN_ACCESSWRITE | NFSV4OPEN_ACCESSREAD); *tl++ = txdr_unsigned(NFSV4OPEN_DENYNONE); tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; (void) nfsm_strtom(nd, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OPEN_CREATE); if (fmode & O_EXCL) { if (NFSHASNFSV4N(nmp)) { if (NFSHASSESSPERSIST(nmp)) { /* Use GUARDED for persistent sessions. */ *tl = txdr_unsigned(NFSCREATE_GUARDED); nfscl_fillsattr(nd, vap, dvp, 0, 0); } else { /* Otherwise, use EXCLUSIVE4_1. */ *tl = txdr_unsigned(NFSCREATE_EXCLUSIVE41); NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); *tl++ = cverf.lval[0]; *tl = cverf.lval[1]; nfscl_fillsattr(nd, vap, dvp, 0, 0); } } else { /* NFSv4.0 */ *tl = txdr_unsigned(NFSCREATE_EXCLUSIVE); NFSM_BUILD(tl, u_int32_t *, NFSX_VERF); *tl++ = cverf.lval[0]; *tl = cverf.lval[1]; } } else { *tl = txdr_unsigned(NFSCREATE_UNCHECKED); nfscl_fillsattr(nd, vap, dvp, 0, 0); } NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OPEN_CLAIMNULL); (void) nfsm_strtom(nd, name, namelen); /* Get the new file's handle and attributes. */ NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OP_GETFH); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); /* Get the directory's post-op attributes. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_PUTFH); (void) nfsm_fhtom(nd, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, 0); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); NFSCL_INCRSEQID(owp->nfsow_seqid, nd); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID + 6 * NFSX_UNSIGNED); stateid.seqid = *tl++; stateid.other[0] = *tl++; stateid.other[1] = *tl++; stateid.other[2] = *tl; rflags = fxdr_unsigned(u_int32_t, *(tl + 6)); (void) nfsrv_getattrbits(nd, &attrbits, NULL, NULL); NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); deleg = fxdr_unsigned(int, *tl); if (deleg == NFSV4OPEN_DELEGATEREAD || deleg == NFSV4OPEN_DELEGATEWRITE) { if (!(owp->nfsow_clp->nfsc_flags & NFSCLFLAGS_FIRSTDELEG)) owp->nfsow_clp->nfsc_flags |= (NFSCLFLAGS_FIRSTDELEG | NFSCLFLAGS_GOTDELEG); MALLOC(dp, struct nfscldeleg *, sizeof (struct nfscldeleg) + NFSX_V4FHMAX, M_NFSCLDELEG, M_WAITOK); LIST_INIT(&dp->nfsdl_owner); LIST_INIT(&dp->nfsdl_lock); dp->nfsdl_clp = owp->nfsow_clp; newnfs_copyincred(cred, &dp->nfsdl_cred); nfscl_lockinit(&dp->nfsdl_rwlock); NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID + NFSX_UNSIGNED); dp->nfsdl_stateid.seqid = *tl++; dp->nfsdl_stateid.other[0] = *tl++; dp->nfsdl_stateid.other[1] = *tl++; dp->nfsdl_stateid.other[2] = *tl++; ret = fxdr_unsigned(int, *tl); if (deleg == NFSV4OPEN_DELEGATEWRITE) { dp->nfsdl_flags = NFSCLDL_WRITE; /* * Indicates how much the file can grow. */ NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); limitby = fxdr_unsigned(int, *tl++); switch (limitby) { case NFSV4OPEN_LIMITSIZE: dp->nfsdl_sizelimit = fxdr_hyper(tl); break; case NFSV4OPEN_LIMITBLOCKS: dp->nfsdl_sizelimit = fxdr_unsigned(u_int64_t, *tl++); dp->nfsdl_sizelimit *= fxdr_unsigned(u_int64_t, *tl); break; default: error = NFSERR_BADXDR; goto nfsmout; } } else { dp->nfsdl_flags = NFSCLDL_READ; } if (ret) dp->nfsdl_flags |= NFSCLDL_RECALL; error = nfsrv_dissectace(nd, &dp->nfsdl_ace, &ret, &acesize, p); if (error) goto nfsmout; } else if (deleg != NFSV4OPEN_DELEGATENONE) { error = NFSERR_BADXDR; goto nfsmout; } error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp); if (error) goto nfsmout; /* Get rid of the PutFH and Getattr status values. */ NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED); /* Load the directory attributes. */ error = nfsm_loadattr(nd, dnap); if (error) goto nfsmout; *dattrflagp = 1; if (dp != NULL && *attrflagp) { dp->nfsdl_change = nnap->na_filerev; dp->nfsdl_modtime = nnap->na_mtime; dp->nfsdl_flags |= NFSCLDL_MODTIMESET; } /* * We can now complete the Open state. */ nfhp = *nfhpp; if (dp != NULL) { dp->nfsdl_fhlen = nfhp->nfh_len; NFSBCOPY(nfhp->nfh_fh, dp->nfsdl_fh, nfhp->nfh_len); } /* * Get an Open structure that will be * attached to the OpenOwner, acquired already. */ error = nfscl_open(dvp, nfhp->nfh_fh, nfhp->nfh_len, (NFSV4OPEN_ACCESSWRITE | NFSV4OPEN_ACCESSREAD), 0, cred, p, NULL, &op, &newone, NULL, 0); if (error) goto nfsmout; op->nfso_stateid = stateid; newnfs_copyincred(cred, &op->nfso_cred); if ((rflags & NFSV4OPEN_RESULTCONFIRM)) { do { ret = nfsrpc_openconfirm(dvp, nfhp->nfh_fh, nfhp->nfh_len, op, cred, p); if (ret == NFSERR_DELAY) (void) nfs_catnap(PZERO, ret, "nfs_create"); } while (ret == NFSERR_DELAY); error = ret; } /* * If the server is handing out delegations, but we didn't * get one because an OpenConfirm was required, try the * Open again, to get a delegation. This is a harmless no-op, * from a server's point of view. */ if ((rflags & NFSV4OPEN_RESULTCONFIRM) && (owp->nfsow_clp->nfsc_flags & NFSCLFLAGS_GOTDELEG) && !error && dp == NULL) { do { ret = nfsrpc_openrpc(VFSTONFS(vnode_mount(dvp)), dvp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, nfhp->nfh_fh, nfhp->nfh_len, (NFSV4OPEN_ACCESSWRITE | NFSV4OPEN_ACCESSREAD), op, name, namelen, &dp, 0, 0x0, cred, p, 0, 1); if (ret == NFSERR_DELAY) (void) nfs_catnap(PZERO, ret, "nfs_crt2"); } while (ret == NFSERR_DELAY); if (ret) { if (dp != NULL) { FREE((caddr_t)dp, M_NFSCLDELEG); dp = NULL; } if (ret == NFSERR_STALECLIENTID || ret == NFSERR_STALEDONTRECOVER || ret == NFSERR_BADSESSION) error = ret; } } nfscl_openrelease(nmp, op, error, newone); *unlockedp = 1; } if (nd->nd_repstat != 0 && error == 0) error = nd->nd_repstat; if (error == NFSERR_STALECLIENTID) nfscl_initiate_recovery(owp->nfsow_clp); nfsmout: if (!error) *dpp = dp; else if (dp != NULL) FREE((caddr_t)dp, M_NFSCLDELEG); mbuf_freem(nd->nd_mrep); return (error); } /* * Nfs remove rpc */ APPLESTATIC int nfsrpc_remove(vnode_t dvp, char *name, int namelen, vnode_t vp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, int *dattrflagp, void *dstuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsnode *np; struct nfsmount *nmp; nfsv4stateid_t dstateid; int error, ret = 0, i; *dattrflagp = 0; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); nmp = VFSTONFS(vnode_mount(dvp)); tryagain: if (NFSHASNFSV4(nmp) && ret == 0) { ret = nfscl_removedeleg(vp, p, &dstateid); if (ret == 1) { NFSCL_REQSTART(nd, NFSPROC_RETDELEGREMOVE, vp); NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID + NFSX_UNSIGNED); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = dstateid.seqid; *tl++ = dstateid.other[0]; *tl++ = dstateid.other[1]; *tl++ = dstateid.other[2]; *tl = txdr_unsigned(NFSV4OP_PUTFH); np = VTONFS(dvp); (void) nfsm_fhtom(nd, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, 0); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_REMOVE); } } else { ret = 0; } if (ret == 0) NFSCL_REQSTART(nd, NFSPROC_REMOVE, dvp); (void) nfsm_strtom(nd, name, namelen); error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { /* For NFSv4, parse out any Delereturn replies. */ if (ret > 0 && nd->nd_repstat != 0 && (nd->nd_flag & ND_NOMOREDATA)) { /* * If the Delegreturn failed, try again without * it. The server will Recall, as required. */ mbuf_freem(nd->nd_mrep); goto tryagain; } for (i = 0; i < (ret * 2); i++) { if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) nd->nd_flag |= ND_NOMOREDATA; } } error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); } if (nd->nd_repstat && !error) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Do an nfs rename rpc. */ APPLESTATIC int nfsrpc_rename(vnode_t fdvp, vnode_t fvp, char *fnameptr, int fnamelen, vnode_t tdvp, vnode_t tvp, char *tnameptr, int tnamelen, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *fnap, struct nfsvattr *tnap, int *fattrflagp, int *tattrflagp, void *fstuff, void *tstuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp; struct nfsnode *np; nfsattrbit_t attrbits; nfsv4stateid_t fdstateid, tdstateid; int error = 0, ret = 0, gottd = 0, gotfd = 0, i; *fattrflagp = 0; *tattrflagp = 0; nmp = VFSTONFS(vnode_mount(fdvp)); if (fnamelen > NFS_MAXNAMLEN || tnamelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); tryagain: if (NFSHASNFSV4(nmp) && ret == 0) { ret = nfscl_renamedeleg(fvp, &fdstateid, &gotfd, tvp, &tdstateid, &gottd, p); if (gotfd && gottd) { NFSCL_REQSTART(nd, NFSPROC_RETDELEGRENAME2, fvp); } else if (gotfd) { NFSCL_REQSTART(nd, NFSPROC_RETDELEGRENAME1, fvp); } else if (gottd) { NFSCL_REQSTART(nd, NFSPROC_RETDELEGRENAME1, tvp); } if (gotfd) { NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = fdstateid.seqid; *tl++ = fdstateid.other[0]; *tl++ = fdstateid.other[1]; *tl = fdstateid.other[2]; if (gottd) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_PUTFH); np = VTONFS(tvp); (void) nfsm_fhtom(nd, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, 0); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_DELEGRETURN); } } if (gottd) { NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = tdstateid.seqid; *tl++ = tdstateid.other[0]; *tl++ = tdstateid.other[1]; *tl = tdstateid.other[2]; } if (ret > 0) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_PUTFH); np = VTONFS(fdvp); (void) nfsm_fhtom(nd, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, 0); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_SAVEFH); } } else { ret = 0; } if (ret == 0) NFSCL_REQSTART(nd, NFSPROC_RENAME, fdvp); if (nd->nd_flag & ND_NFSV4) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSWCCATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_PUTFH); (void) nfsm_fhtom(nd, VTONFS(tdvp)->n_fhp->nfh_fh, VTONFS(tdvp)->n_fhp->nfh_len, 0); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); nd->nd_flag |= ND_V4WCCATTR; NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_RENAME); } (void) nfsm_strtom(nd, fnameptr, fnamelen); if (!(nd->nd_flag & ND_NFSV4)) (void) nfsm_fhtom(nd, VTONFS(tdvp)->n_fhp->nfh_fh, VTONFS(tdvp)->n_fhp->nfh_len, 0); (void) nfsm_strtom(nd, tnameptr, tnamelen); error = nfscl_request(nd, fdvp, p, cred, fstuff); if (error) return (error); if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) { /* For NFSv4, parse out any Delereturn replies. */ if (ret > 0 && nd->nd_repstat != 0 && (nd->nd_flag & ND_NOMOREDATA)) { /* * If the Delegreturn failed, try again without * it. The server will Recall, as required. */ mbuf_freem(nd->nd_mrep); goto tryagain; } for (i = 0; i < (ret * 2); i++) { if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) { if (i == 0 && ret > 1) { /* * If the Delegreturn failed, try again * without it. The server will Recall, as * required. * If ret > 1, the first iteration of this * loop is the second DelegReturn result. */ mbuf_freem(nd->nd_mrep); goto tryagain; } else { nd->nd_flag |= ND_NOMOREDATA; } } } } /* Now, the first wcc attribute reply. */ if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) nd->nd_flag |= ND_NOMOREDATA; } error = nfscl_wcc_data(nd, fdvp, fnap, fattrflagp, NULL, fstuff); /* and the second wcc attribute reply. */ if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4 && !error) { NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) nd->nd_flag |= ND_NOMOREDATA; } if (!error) error = nfscl_wcc_data(nd, tdvp, tnap, tattrflagp, NULL, tstuff); } if (nd->nd_repstat && !error) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs hard link create rpc */ APPLESTATIC int nfsrpc_link(vnode_t dvp, vnode_t vp, char *name, int namelen, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nap, int *attrflagp, int *dattrflagp, void *dstuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; nfsattrbit_t attrbits; int error = 0; *attrflagp = 0; *dattrflagp = 0; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_LINK, vp); if (nd->nd_flag & ND_NFSV4) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_PUTFH); } (void) nfsm_fhtom(nd, VTONFS(dvp)->n_fhp->nfh_fh, VTONFS(dvp)->n_fhp->nfh_len, 0); if (nd->nd_flag & ND_NFSV4) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSWCCATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); nd->nd_flag |= ND_V4WCCATTR; NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_LINK); } (void) nfsm_strtom(nd, name, namelen); error = nfscl_request(nd, vp, p, cred, dstuff); if (error) return (error); if (nd->nd_flag & ND_NFSV3) { error = nfscl_postop_attr(nd, nap, attrflagp, dstuff); if (!error) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); } else if ((nd->nd_flag & (ND_NFSV4 | ND_NOMOREDATA)) == ND_NFSV4) { /* * First, parse out the PutFH and Getattr result. */ NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (!(*(tl + 1))) NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED); if (*(tl + 1)) nd->nd_flag |= ND_NOMOREDATA; /* * Get the pre-op attributes. */ error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); } if (nd->nd_repstat && !error) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs symbolic link create rpc */ APPLESTATIC int nfsrpc_symlink(vnode_t dvp, char *name, int namelen, char *target, struct vattr *vap, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp; int slen, error = 0; *nfhpp = NULL; *attrflagp = 0; *dattrflagp = 0; nmp = VFSTONFS(vnode_mount(dvp)); slen = strlen(target); if (slen > NFS_MAXPATHLEN || namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_SYMLINK, dvp); if (nd->nd_flag & ND_NFSV4) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFLNK); (void) nfsm_strtom(nd, target, slen); } (void) nfsm_strtom(nd, name, namelen); if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) nfscl_fillsattr(nd, vap, dvp, 0, 0); if (!(nd->nd_flag & ND_NFSV4)) (void) nfsm_strtom(nd, target, slen); if (nd->nd_flag & ND_NFSV2) nfscl_fillsattr(nd, vap, dvp, NFSSATTR_SIZENEG1, 0); error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); if (nd->nd_flag & ND_NFSV4) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if ((nd->nd_flag & ND_NFSV3) && !error) { if (!nd->nd_repstat) error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp); if (!error) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); } if (nd->nd_repstat && !error) error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); /* * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. * Only do this if vfs.nfs.ignore_eexist is set. * Never do this for NFSv4.1 or later minor versions, since sessions * should guarantee "exactly once" RPC semantics. */ if (error == EEXIST && nfsignore_eexist != 0 && (!NFSHASNFSV4(nmp) || nmp->nm_minorvers == 0)) error = 0; return (error); } /* * nfs make dir rpc */ APPLESTATIC int nfsrpc_mkdir(vnode_t dvp, char *name, int namelen, struct vattr *vap, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, struct nfsvattr *nnap, struct nfsfh **nfhpp, int *attrflagp, int *dattrflagp, void *dstuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; nfsattrbit_t attrbits; int error = 0; struct nfsfh *fhp; struct nfsmount *nmp; *nfhpp = NULL; *attrflagp = 0; *dattrflagp = 0; nmp = VFSTONFS(vnode_mount(dvp)); fhp = VTONFS(dvp)->n_fhp; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_MKDIR, dvp); if (nd->nd_flag & ND_NFSV4) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFDIR); } (void) nfsm_strtom(nd, name, namelen); nfscl_fillsattr(nd, vap, dvp, NFSSATTR_SIZENEG1, 0); if (nd->nd_flag & ND_NFSV4) { NFSGETATTR_ATTRBIT(&attrbits); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OP_GETFH); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_PUTFH); (void) nfsm_fhtom(nd, fhp->nfh_fh, fhp->nfh_len, 0); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); if (nd->nd_flag & ND_NFSV4) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if (!nd->nd_repstat && !error) { if (nd->nd_flag & ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL); } if (!error) error = nfscl_mtofh(nd, nfhpp, nnap, attrflagp); if (error == 0 && (nd->nd_flag & ND_NFSV4) != 0) { /* Get rid of the PutFH and Getattr status values. */ NFSM_DISSECT(tl, u_int32_t *, 4 * NFSX_UNSIGNED); /* Load the directory attributes. */ error = nfsm_loadattr(nd, dnap); if (error == 0) *dattrflagp = 1; } } if ((nd->nd_flag & ND_NFSV3) && !error) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if (nd->nd_repstat && !error) error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); /* * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry. * Only do this if vfs.nfs.ignore_eexist is set. * Never do this for NFSv4.1 or later minor versions, since sessions * should guarantee "exactly once" RPC semantics. */ if (error == EEXIST && nfsignore_eexist != 0 && (!NFSHASNFSV4(nmp) || nmp->nm_minorvers == 0)) error = 0; return (error); } /* * nfs remove directory call */ APPLESTATIC int nfsrpc_rmdir(vnode_t dvp, char *name, int namelen, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *dnap, int *dattrflagp, void *dstuff) { struct nfsrv_descript nfsd, *nd = &nfsd; int error = 0; *dattrflagp = 0; if (namelen > NFS_MAXNAMLEN) return (ENAMETOOLONG); NFSCL_REQSTART(nd, NFSPROC_RMDIR, dvp); (void) nfsm_strtom(nd, name, namelen); error = nfscl_request(nd, dvp, p, cred, dstuff); if (error) return (error); if (nd->nd_flag & (ND_NFSV3 | ND_NFSV4)) error = nfscl_wcc_data(nd, dvp, dnap, dattrflagp, NULL, dstuff); if (nd->nd_repstat && !error) error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); /* * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry. */ if (error == ENOENT) error = 0; return (error); } /* * Readdir rpc. * Always returns with either uio_resid unchanged, if you are at the * end of the directory, or uio_resid == 0, with all DIRBLKSIZ chunks * filled in. * I felt this would allow caching of directory blocks more easily * than returning a pertially filled block. * Directory offset cookies: * Oh my, what to do with them... * I can think of three ways to deal with them: * 1 - have the layer above these RPCs maintain a map between logical * directory byte offsets and the NFS directory offset cookies * 2 - pass the opaque directory offset cookies up into userland * and let the libc functions deal with them, via the system call * 3 - return them to userland in the "struct dirent", so future versions * of libc can use them and do whatever is necessary to make things work * above these rpc calls, in the meantime * For now, I do #3 by "hiding" the directory offset cookies after the * d_name field in struct dirent. This is space inside d_reclen that * will be ignored by anything that doesn't know about them. * The directory offset cookies are filled in as the last 8 bytes of * each directory entry, after d_name. Someday, the userland libc * functions may be able to use these. In the meantime, it satisfies * OpenBSD's requirements for cookies being returned. * If expects the directory offset cookie for the read to be in uio_offset * and returns the one for the next entry after this directory block in * there, as well. */ APPLESTATIC int nfsrpc_readdir(vnode_t vp, struct uio *uiop, nfsuint64 *cookiep, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, int *eofp, void *stuff) { int len, left; struct dirent *dp = NULL; u_int32_t *tl; nfsquad_t cookie, ncookie; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsnode *dnp = VTONFS(vp); struct nfsvattr nfsva; struct nfsrv_descript nfsd, *nd = &nfsd; int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; int reqsize, tryformoredirs = 1, readsize, eof = 0, gotmnton = 0; long dotfileid, dotdotfileid = 0; u_int32_t fakefileno = 0xffffffff, rderr; char *cp; nfsattrbit_t attrbits, dattrbits; u_int32_t *tl2 = NULL; size_t tresid; KASSERT(uiop->uio_iovcnt == 1 && (uio_uio_resid(uiop) & (DIRBLKSIZ - 1)) == 0, ("nfs readdirrpc bad uio")); /* * There is no point in reading a lot more than uio_resid, however * adding one additional DIRBLKSIZ makes sense. Since uio_resid * and nm_readdirsize are both exact multiples of DIRBLKSIZ, this * will never make readsize > nm_readdirsize. */ readsize = nmp->nm_readdirsize; if (readsize > uio_uio_resid(uiop)) readsize = uio_uio_resid(uiop) + DIRBLKSIZ; *attrflagp = 0; if (eofp) *eofp = 0; tresid = uio_uio_resid(uiop); cookie.lval[0] = cookiep->nfsuquad[0]; cookie.lval[1] = cookiep->nfsuquad[1]; nd->nd_mrep = NULL; /* * For NFSv4, first create the "." and ".." entries. */ if (NFSHASNFSV4(nmp)) { reqsize = 6 * NFSX_UNSIGNED; NFSGETATTR_ATTRBIT(&dattrbits); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FILEID); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TYPE); if (NFSISSET_ATTRBIT(&dnp->n_vattr.na_suppattr, NFSATTRBIT_MOUNTEDONFILEID)) { NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MOUNTEDONFILEID); gotmnton = 1; } else { /* * Must fake it. Use the fileno, except when the * fsid is != to that of the directory. For that * case, generate a fake fileno that is not the same. */ NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FSID); gotmnton = 0; } /* * Joy, oh joy. For V4 we get to hand craft '.' and '..'. */ if (uiop->uio_offset == 0) { NFSCL_REQSTART(nd, NFSPROC_LOOKUPP, vp); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OP_GETFH); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); dotfileid = 0; /* Fake out the compiler. */ if ((nd->nd_flag & ND_NOMOREDATA) == 0) { error = nfsm_loadattr(nd, &nfsva); if (error != 0) goto nfsmout; dotfileid = nfsva.na_fileid; } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); len = fxdr_unsigned(int, *(tl + 4)); if (len > 0 && len <= NFSX_V4FHMAX) error = nfsm_advance(nd, NFSM_RNDUP(len), -1); else error = EPERM; if (!error) { NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED); nfsva.na_mntonfileno = 0xffffffff; error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, p, cred); if (error) { dotdotfileid = dotfileid; } else if (gotmnton) { if (nfsva.na_mntonfileno != 0xffffffff) dotdotfileid = nfsva.na_mntonfileno; else dotdotfileid = nfsva.na_fileid; } else if (nfsva.na_filesid[0] == dnp->n_vattr.na_filesid[0] && nfsva.na_filesid[1] == dnp->n_vattr.na_filesid[1]) { dotdotfileid = nfsva.na_fileid; } else { do { fakefileno--; } while (fakefileno == nfsva.na_fileid); dotdotfileid = fakefileno; } } } else if (nd->nd_repstat == NFSERR_NOENT) { /* * Lookupp returns NFSERR_NOENT when we are * at the root, so just use the current dir. */ nd->nd_repstat = 0; dotdotfileid = dotfileid; } else { error = nd->nd_repstat; } mbuf_freem(nd->nd_mrep); if (error) return (error); nd->nd_mrep = NULL; dp = (struct dirent *)uio_iov_base(uiop); dp->d_off = 0; dp->d_type = DT_DIR; dp->d_fileno = dotfileid; dp->d_namlen = 1; *((uint64_t *)dp->d_name) = 0; /* Zero pad it. */ dp->d_name[0] = '.'; dp->d_reclen = _GENERIC_DIRSIZ(dp) + NFSX_HYPER; /* * Just make these offset cookie 0. */ tl = (u_int32_t *)&dp->d_name[8]; *tl++ = 0; *tl = 0; blksiz += dp->d_reclen; uio_uio_resid_add(uiop, -(dp->d_reclen)); uiop->uio_offset += dp->d_reclen; uio_iov_base_add(uiop, dp->d_reclen); uio_iov_len_add(uiop, -(dp->d_reclen)); dp = (struct dirent *)uio_iov_base(uiop); dp->d_off = 0; dp->d_type = DT_DIR; dp->d_fileno = dotdotfileid; dp->d_namlen = 2; *((uint64_t *)dp->d_name) = 0; dp->d_name[0] = '.'; dp->d_name[1] = '.'; dp->d_reclen = _GENERIC_DIRSIZ(dp) + NFSX_HYPER; /* * Just make these offset cookie 0. */ tl = (u_int32_t *)&dp->d_name[8]; *tl++ = 0; *tl = 0; blksiz += dp->d_reclen; uio_uio_resid_add(uiop, -(dp->d_reclen)); uiop->uio_offset += dp->d_reclen; uio_iov_base_add(uiop, dp->d_reclen); uio_iov_len_add(uiop, -(dp->d_reclen)); } NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_RDATTRERROR); } else { reqsize = 5 * NFSX_UNSIGNED; } /* * Loop around doing readdir rpc's of size readsize. * The stopping criteria is EOF or buffer full. */ while (more_dirs && bigenough) { *attrflagp = 0; NFSCL_REQSTART(nd, NFSPROC_READDIR, vp); if (nd->nd_flag & ND_NFSV2) { NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = cookie.lval[1]; *tl = txdr_unsigned(readsize); } else { NFSM_BUILD(tl, u_int32_t *, reqsize); *tl++ = cookie.lval[0]; *tl++ = cookie.lval[1]; if (cookie.qval == 0) { *tl++ = 0; *tl++ = 0; } else { NFSLOCKNODE(dnp); *tl++ = dnp->n_cookieverf.nfsuquad[0]; *tl++ = dnp->n_cookieverf.nfsuquad[1]; NFSUNLOCKNODE(dnp); } if (nd->nd_flag & ND_NFSV4) { *tl++ = txdr_unsigned(readsize); *tl = txdr_unsigned(readsize); (void) nfsrv_putattrbit(nd, &attrbits); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &dattrbits); } else { *tl = txdr_unsigned(readsize); } } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (!(nd->nd_flag & ND_NFSV2)) { if (nd->nd_flag & ND_NFSV3) error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (!nd->nd_repstat && !error) { NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER); NFSLOCKNODE(dnp); dnp->n_cookieverf.nfsuquad[0] = *tl++; dnp->n_cookieverf.nfsuquad[1] = *tl; NFSUNLOCKNODE(dnp); } } if (nd->nd_repstat || error) { if (!error) error = nd->nd_repstat; goto nfsmout; } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); more_dirs = fxdr_unsigned(int, *tl); if (!more_dirs) tryformoredirs = 0; /* loop through the dir entries, doctoring them to 4bsd form */ while (more_dirs && bigenough) { if (nd->nd_flag & ND_NFSV4) { NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED); ncookie.lval[0] = *tl++; ncookie.lval[1] = *tl++; len = fxdr_unsigned(int, *tl); } else if (nd->nd_flag & ND_NFSV3) { NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED); nfsva.na_fileid = fxdr_hyper(tl); tl += 2; len = fxdr_unsigned(int, *tl); } else { NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED); nfsva.na_fileid = fxdr_unsigned(long, *tl++); len = fxdr_unsigned(int, *tl); } if (len <= 0 || len > NFS_MAXNAMLEN) { error = EBADRPC; goto nfsmout; } tlen = roundup2(len, 8); if (tlen == len) tlen += 8; /* To ensure null termination. */ left = DIRBLKSIZ - blksiz; if (_GENERIC_DIRLEN(len) + NFSX_HYPER > left) { dp->d_reclen += left; uio_iov_base_add(uiop, left); uio_iov_len_add(uiop, -(left)); uio_uio_resid_add(uiop, -(left)); uiop->uio_offset += left; blksiz = 0; } if (_GENERIC_DIRLEN(len) + NFSX_HYPER > uio_uio_resid(uiop)) bigenough = 0; if (bigenough) { dp = (struct dirent *)uio_iov_base(uiop); dp->d_off = 0; dp->d_namlen = len; dp->d_reclen = _GENERIC_DIRLEN(len) + NFSX_HYPER; dp->d_type = DT_UNKNOWN; blksiz += dp->d_reclen; if (blksiz == DIRBLKSIZ) blksiz = 0; uio_uio_resid_add(uiop, -(DIRHDSIZ)); uiop->uio_offset += DIRHDSIZ; uio_iov_base_add(uiop, DIRHDSIZ); uio_iov_len_add(uiop, -(DIRHDSIZ)); error = nfsm_mbufuio(nd, uiop, len); if (error) goto nfsmout; cp = uio_iov_base(uiop); tlen -= len; *cp = '\0'; /* null terminate */ cp += tlen; /* points to cookie storage */ tl2 = (u_int32_t *)cp; uio_iov_base_add(uiop, (tlen + NFSX_HYPER)); uio_iov_len_add(uiop, -(tlen + NFSX_HYPER)); uio_uio_resid_add(uiop, -(tlen + NFSX_HYPER)); uiop->uio_offset += (tlen + NFSX_HYPER); } else { error = nfsm_advance(nd, NFSM_RNDUP(len), -1); if (error) goto nfsmout; } if (nd->nd_flag & ND_NFSV4) { rderr = 0; nfsva.na_mntonfileno = 0xffffffff; error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, &rderr, p, cred); if (error) goto nfsmout; NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); } else if (nd->nd_flag & ND_NFSV3) { NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED); ncookie.lval[0] = *tl++; ncookie.lval[1] = *tl++; } else { NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED); ncookie.lval[0] = 0; ncookie.lval[1] = *tl++; } if (bigenough) { if (nd->nd_flag & ND_NFSV4) { if (rderr) { dp->d_fileno = 0; } else { if (gotmnton) { if (nfsva.na_mntonfileno != 0xffffffff) dp->d_fileno = nfsva.na_mntonfileno; else dp->d_fileno = nfsva.na_fileid; } else if (nfsva.na_filesid[0] == dnp->n_vattr.na_filesid[0] && nfsva.na_filesid[1] == dnp->n_vattr.na_filesid[1]) { dp->d_fileno = nfsva.na_fileid; } else { do { fakefileno--; } while (fakefileno == nfsva.na_fileid); dp->d_fileno = fakefileno; } dp->d_type = vtonfs_dtype(nfsva.na_type); } } else { dp->d_fileno = nfsva.na_fileid; } *tl2++ = cookiep->nfsuquad[0] = cookie.lval[0] = ncookie.lval[0]; *tl2 = cookiep->nfsuquad[1] = cookie.lval[1] = ncookie.lval[1]; } more_dirs = fxdr_unsigned(int, *tl); } /* * If at end of rpc data, get the eof boolean */ if (!more_dirs) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); eof = fxdr_unsigned(int, *tl); if (tryformoredirs) more_dirs = !eof; if (nd->nd_flag & ND_NFSV4) { error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (error) goto nfsmout; } } mbuf_freem(nd->nd_mrep); nd->nd_mrep = NULL; } /* * Fill last record, iff any, out to a multiple of DIRBLKSIZ * by increasing d_reclen for the last record. */ if (blksiz > 0) { left = DIRBLKSIZ - blksiz; dp->d_reclen += left; uio_iov_base_add(uiop, left); uio_iov_len_add(uiop, -(left)); uio_uio_resid_add(uiop, -(left)); uiop->uio_offset += left; } /* * If returning no data, assume end of file. * If not bigenough, return not end of file, since you aren't * returning all the data * Otherwise, return the eof flag from the server. */ if (eofp) { if (tresid == ((size_t)(uio_uio_resid(uiop)))) *eofp = 1; else if (!bigenough) *eofp = 0; else *eofp = eof; } /* * Add extra empty records to any remaining DIRBLKSIZ chunks. */ while (uio_uio_resid(uiop) > 0 && uio_uio_resid(uiop) != tresid) { dp = (struct dirent *)uio_iov_base(uiop); dp->d_type = DT_UNKNOWN; dp->d_fileno = 0; dp->d_namlen = 0; dp->d_name[0] = '\0'; tl = (u_int32_t *)&dp->d_name[4]; *tl++ = cookie.lval[0]; *tl = cookie.lval[1]; dp->d_reclen = DIRBLKSIZ; uio_iov_base_add(uiop, DIRBLKSIZ); uio_iov_len_add(uiop, -(DIRBLKSIZ)); uio_uio_resid_add(uiop, -(DIRBLKSIZ)); uiop->uio_offset += DIRBLKSIZ; } nfsmout: if (nd->nd_mrep != NULL) mbuf_freem(nd->nd_mrep); return (error); } #ifndef APPLE /* * NFS V3 readdir plus RPC. Used in place of nfsrpc_readdir(). * (Also used for NFS V4 when mount flag set.) * (ditto above w.r.t. multiple of DIRBLKSIZ, etc.) */ APPLESTATIC int nfsrpc_readdirplus(vnode_t vp, struct uio *uiop, nfsuint64 *cookiep, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, int *eofp, void *stuff) { int len, left; struct dirent *dp = NULL; u_int32_t *tl; vnode_t newvp = NULLVP; struct nfsrv_descript nfsd, *nd = &nfsd; struct nameidata nami, *ndp = &nami; struct componentname *cnp = &ndp->ni_cnd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsnode *dnp = VTONFS(vp), *np; struct nfsvattr nfsva; struct nfsfh *nfhp; nfsquad_t cookie, ncookie; int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1; int attrflag, tryformoredirs = 1, eof = 0, gotmnton = 0; int isdotdot = 0, unlocknewvp = 0; long dotfileid, dotdotfileid = 0, fileno = 0; char *cp; nfsattrbit_t attrbits, dattrbits; size_t tresid; u_int32_t *tl2 = NULL, fakefileno = 0xffffffff, rderr; struct timespec dctime; KASSERT(uiop->uio_iovcnt == 1 && (uio_uio_resid(uiop) & (DIRBLKSIZ - 1)) == 0, ("nfs readdirplusrpc bad uio")); timespecclear(&dctime); *attrflagp = 0; if (eofp != NULL) *eofp = 0; ndp->ni_dvp = vp; nd->nd_mrep = NULL; cookie.lval[0] = cookiep->nfsuquad[0]; cookie.lval[1] = cookiep->nfsuquad[1]; tresid = uio_uio_resid(uiop); /* * For NFSv4, first create the "." and ".." entries. */ if (NFSHASNFSV4(nmp)) { NFSGETATTR_ATTRBIT(&dattrbits); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FILEID); if (NFSISSET_ATTRBIT(&dnp->n_vattr.na_suppattr, NFSATTRBIT_MOUNTEDONFILEID)) { NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MOUNTEDONFILEID); gotmnton = 1; } else { /* * Must fake it. Use the fileno, except when the * fsid is != to that of the directory. For that * case, generate a fake fileno that is not the same. */ NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_FSID); gotmnton = 0; } /* * Joy, oh joy. For V4 we get to hand craft '.' and '..'. */ if (uiop->uio_offset == 0) { NFSCL_REQSTART(nd, NFSPROC_LOOKUPP, vp); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFSV4OP_GETFH); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &attrbits); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); dotfileid = 0; /* Fake out the compiler. */ if ((nd->nd_flag & ND_NOMOREDATA) == 0) { error = nfsm_loadattr(nd, &nfsva); if (error != 0) goto nfsmout; dctime = nfsva.na_ctime; dotfileid = nfsva.na_fileid; } if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED); len = fxdr_unsigned(int, *(tl + 4)); if (len > 0 && len <= NFSX_V4FHMAX) error = nfsm_advance(nd, NFSM_RNDUP(len), -1); else error = EPERM; if (!error) { NFSM_DISSECT(tl, u_int32_t *, 2*NFSX_UNSIGNED); nfsva.na_mntonfileno = 0xffffffff; error = nfsv4_loadattr(nd, NULL, &nfsva, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, p, cred); if (error) { dotdotfileid = dotfileid; } else if (gotmnton) { if (nfsva.na_mntonfileno != 0xffffffff) dotdotfileid = nfsva.na_mntonfileno; else dotdotfileid = nfsva.na_fileid; } else if (nfsva.na_filesid[0] == dnp->n_vattr.na_filesid[0] && nfsva.na_filesid[1] == dnp->n_vattr.na_filesid[1]) { dotdotfileid = nfsva.na_fileid; } else { do { fakefileno--; } while (fakefileno == nfsva.na_fileid); dotdotfileid = fakefileno; } } } else if (nd->nd_repstat == NFSERR_NOENT) { /* * Lookupp returns NFSERR_NOENT when we are * at the root, so just use the current dir. */ nd->nd_repstat = 0; dotdotfileid = dotfileid; } else { error = nd->nd_repstat; } mbuf_freem(nd->nd_mrep); if (error) return (error); nd->nd_mrep = NULL; dp = (struct dirent *)uio_iov_base(uiop); dp->d_off = 0; dp->d_type = DT_DIR; dp->d_fileno = dotfileid; dp->d_namlen = 1; *((uint64_t *)dp->d_name) = 0; /* Zero pad it. */ dp->d_name[0] = '.'; dp->d_reclen = _GENERIC_DIRSIZ(dp) + NFSX_HYPER; /* * Just make these offset cookie 0. */ tl = (u_int32_t *)&dp->d_name[8]; *tl++ = 0; *tl = 0; blksiz += dp->d_reclen; uio_uio_resid_add(uiop, -(dp->d_reclen)); uiop->uio_offset += dp->d_reclen; uio_iov_base_add(uiop, dp->d_reclen); uio_iov_len_add(uiop, -(dp->d_reclen)); dp = (struct dirent *)uio_iov_base(uiop); dp->d_off = 0; dp->d_type = DT_DIR; dp->d_fileno = dotdotfileid; dp->d_namlen = 2; *((uint64_t *)dp->d_name) = 0; dp->d_name[0] = '.'; dp->d_name[1] = '.'; dp->d_reclen = _GENERIC_DIRSIZ(dp) + NFSX_HYPER; /* * Just make these offset cookie 0. */ tl = (u_int32_t *)&dp->d_name[8]; *tl++ = 0; *tl = 0; blksiz += dp->d_reclen; uio_uio_resid_add(uiop, -(dp->d_reclen)); uiop->uio_offset += dp->d_reclen; uio_iov_base_add(uiop, dp->d_reclen); uio_iov_len_add(uiop, -(dp->d_reclen)); } NFSREADDIRPLUS_ATTRBIT(&attrbits); if (gotmnton) NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MOUNTEDONFILEID); } /* * Loop around doing readdir rpc's of size nm_readdirsize. * The stopping criteria is EOF or buffer full. */ while (more_dirs && bigenough) { *attrflagp = 0; NFSCL_REQSTART(nd, NFSPROC_READDIRPLUS, vp); NFSM_BUILD(tl, u_int32_t *, 6 * NFSX_UNSIGNED); *tl++ = cookie.lval[0]; *tl++ = cookie.lval[1]; if (cookie.qval == 0) { *tl++ = 0; *tl++ = 0; } else { NFSLOCKNODE(dnp); *tl++ = dnp->n_cookieverf.nfsuquad[0]; *tl++ = dnp->n_cookieverf.nfsuquad[1]; NFSUNLOCKNODE(dnp); } *tl++ = txdr_unsigned(nmp->nm_readdirsize); *tl = txdr_unsigned(nmp->nm_readdirsize); if (nd->nd_flag & ND_NFSV4) { (void) nfsrv_putattrbit(nd, &attrbits); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); (void) nfsrv_putattrbit(nd, &dattrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_flag & ND_NFSV3) error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (nd->nd_repstat || error) { if (!error) error = nd->nd_repstat; goto nfsmout; } if ((nd->nd_flag & ND_NFSV3) != 0 && *attrflagp != 0) dctime = nap->na_ctime; NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); NFSLOCKNODE(dnp); dnp->n_cookieverf.nfsuquad[0] = *tl++; dnp->n_cookieverf.nfsuquad[1] = *tl++; NFSUNLOCKNODE(dnp); more_dirs = fxdr_unsigned(int, *tl); if (!more_dirs) tryformoredirs = 0; /* loop through the dir entries, doctoring them to 4bsd form */ while (more_dirs && bigenough) { NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED); if (nd->nd_flag & ND_NFSV4) { ncookie.lval[0] = *tl++; ncookie.lval[1] = *tl++; } else { fileno = fxdr_unsigned(long, *++tl); tl++; } len = fxdr_unsigned(int, *tl); if (len <= 0 || len > NFS_MAXNAMLEN) { error = EBADRPC; goto nfsmout; } tlen = roundup2(len, 8); if (tlen == len) tlen += 8; /* To ensure null termination. */ left = DIRBLKSIZ - blksiz; if (_GENERIC_DIRLEN(len) + NFSX_HYPER > left) { dp->d_reclen += left; uio_iov_base_add(uiop, left); uio_iov_len_add(uiop, -(left)); uio_uio_resid_add(uiop, -(left)); uiop->uio_offset += left; blksiz = 0; } if (_GENERIC_DIRLEN(len) + NFSX_HYPER > uio_uio_resid(uiop)) bigenough = 0; if (bigenough) { dp = (struct dirent *)uio_iov_base(uiop); dp->d_off = 0; dp->d_namlen = len; dp->d_reclen = _GENERIC_DIRLEN(len) + NFSX_HYPER; dp->d_type = DT_UNKNOWN; blksiz += dp->d_reclen; if (blksiz == DIRBLKSIZ) blksiz = 0; uio_uio_resid_add(uiop, -(DIRHDSIZ)); uiop->uio_offset += DIRHDSIZ; uio_iov_base_add(uiop, DIRHDSIZ); uio_iov_len_add(uiop, -(DIRHDSIZ)); cnp->cn_nameptr = uio_iov_base(uiop); cnp->cn_namelen = len; NFSCNHASHZERO(cnp); error = nfsm_mbufuio(nd, uiop, len); if (error) goto nfsmout; cp = uio_iov_base(uiop); tlen -= len; *cp = '\0'; cp += tlen; /* points to cookie storage */ tl2 = (u_int32_t *)cp; if (len == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') isdotdot = 1; else isdotdot = 0; uio_iov_base_add(uiop, (tlen + NFSX_HYPER)); uio_iov_len_add(uiop, -(tlen + NFSX_HYPER)); uio_uio_resid_add(uiop, -(tlen + NFSX_HYPER)); uiop->uio_offset += (tlen + NFSX_HYPER); } else { error = nfsm_advance(nd, NFSM_RNDUP(len), -1); if (error) goto nfsmout; } nfhp = NULL; if (nd->nd_flag & ND_NFSV3) { NFSM_DISSECT(tl, u_int32_t *, 3*NFSX_UNSIGNED); ncookie.lval[0] = *tl++; ncookie.lval[1] = *tl++; attrflag = fxdr_unsigned(int, *tl); if (attrflag) { error = nfsm_loadattr(nd, &nfsva); if (error) goto nfsmout; } NFSM_DISSECT(tl,u_int32_t *,NFSX_UNSIGNED); if (*tl) { error = nfsm_getfh(nd, &nfhp); if (error) goto nfsmout; } if (!attrflag && nfhp != NULL) { FREE((caddr_t)nfhp, M_NFSFH); nfhp = NULL; } } else { rderr = 0; nfsva.na_mntonfileno = 0xffffffff; error = nfsv4_loadattr(nd, NULL, &nfsva, &nfhp, NULL, 0, NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, &rderr, p, cred); if (error) goto nfsmout; } if (bigenough) { if (nd->nd_flag & ND_NFSV4) { if (rderr) { dp->d_fileno = 0; } else if (gotmnton) { if (nfsva.na_mntonfileno != 0xffffffff) dp->d_fileno = nfsva.na_mntonfileno; else dp->d_fileno = nfsva.na_fileid; } else if (nfsva.na_filesid[0] == dnp->n_vattr.na_filesid[0] && nfsva.na_filesid[1] == dnp->n_vattr.na_filesid[1]) { dp->d_fileno = nfsva.na_fileid; } else { do { fakefileno--; } while (fakefileno == nfsva.na_fileid); dp->d_fileno = fakefileno; } } else { dp->d_fileno = fileno; } *tl2++ = cookiep->nfsuquad[0] = cookie.lval[0] = ncookie.lval[0]; *tl2 = cookiep->nfsuquad[1] = cookie.lval[1] = ncookie.lval[1]; if (nfhp != NULL) { if (NFSRV_CMPFH(nfhp->nfh_fh, nfhp->nfh_len, dnp->n_fhp->nfh_fh, dnp->n_fhp->nfh_len)) { VREF(vp); newvp = vp; unlocknewvp = 0; FREE((caddr_t)nfhp, M_NFSFH); np = dnp; } else if (isdotdot != 0) { /* * Skip doing a nfscl_nget() call for "..". * There's a race between acquiring the nfs * node here and lookups that look for the * directory being read (in the parent). * It would try to get a lock on ".." here, * owning the lock on the directory being * read. Lookup will hold the lock on ".." * and try to acquire the lock on the * directory being read. * If the directory is unlocked/relocked, * then there is a LOR with the buflock * vp is relocked. */ free(nfhp, M_NFSFH); } else { error = nfscl_nget(vnode_mount(vp), vp, nfhp, cnp, p, &np, NULL, LK_EXCLUSIVE); if (!error) { newvp = NFSTOV(np); unlocknewvp = 1; } } nfhp = NULL; if (newvp != NULLVP) { error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL, 0, 0); if (error) { if (unlocknewvp) vput(newvp); else vrele(newvp); goto nfsmout; } dp->d_type = vtonfs_dtype(np->n_vattr.na_type); ndp->ni_vp = newvp; NFSCNHASH(cnp, HASHINIT); if (cnp->cn_namelen <= NCHNAMLEN && (newvp->v_type != VDIR || dctime.tv_sec != 0)) { cache_enter_time(ndp->ni_dvp, ndp->ni_vp, cnp, &nfsva.na_ctime, newvp->v_type != VDIR ? NULL : &dctime); } if (unlocknewvp) vput(newvp); else vrele(newvp); newvp = NULLVP; } } } else if (nfhp != NULL) { FREE((caddr_t)nfhp, M_NFSFH); } NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); more_dirs = fxdr_unsigned(int, *tl); } /* * If at end of rpc data, get the eof boolean */ if (!more_dirs) { NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); eof = fxdr_unsigned(int, *tl); if (tryformoredirs) more_dirs = !eof; if (nd->nd_flag & ND_NFSV4) { error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (error) goto nfsmout; } } mbuf_freem(nd->nd_mrep); nd->nd_mrep = NULL; } /* * Fill last record, iff any, out to a multiple of DIRBLKSIZ * by increasing d_reclen for the last record. */ if (blksiz > 0) { left = DIRBLKSIZ - blksiz; dp->d_reclen += left; uio_iov_base_add(uiop, left); uio_iov_len_add(uiop, -(left)); uio_uio_resid_add(uiop, -(left)); uiop->uio_offset += left; } /* * If returning no data, assume end of file. * If not bigenough, return not end of file, since you aren't * returning all the data * Otherwise, return the eof flag from the server. */ if (eofp != NULL) { if (tresid == uio_uio_resid(uiop)) *eofp = 1; else if (!bigenough) *eofp = 0; else *eofp = eof; } /* * Add extra empty records to any remaining DIRBLKSIZ chunks. */ while (uio_uio_resid(uiop) > 0 && uio_uio_resid(uiop) != tresid) { dp = (struct dirent *)uio_iov_base(uiop); dp->d_type = DT_UNKNOWN; dp->d_fileno = 0; dp->d_namlen = 0; dp->d_name[0] = '\0'; tl = (u_int32_t *)&dp->d_name[4]; *tl++ = cookie.lval[0]; *tl = cookie.lval[1]; dp->d_reclen = DIRBLKSIZ; uio_iov_base_add(uiop, DIRBLKSIZ); uio_iov_len_add(uiop, -(DIRBLKSIZ)); uio_uio_resid_add(uiop, -(DIRBLKSIZ)); uiop->uio_offset += DIRBLKSIZ; } nfsmout: if (nd->nd_mrep != NULL) mbuf_freem(nd->nd_mrep); return (error); } #endif /* !APPLE */ /* * Nfs commit rpc */ APPLESTATIC int nfsrpc_commit(vnode_t vp, u_quad_t offset, int cnt, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; nfsattrbit_t attrbits; int error; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); *attrflagp = 0; NFSCL_REQSTART(nd, NFSPROC_COMMIT, vp); NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED); txdr_hyper(offset, tl); tl += 2; *tl = txdr_unsigned(cnt); if (nd->nd_flag & ND_NFSV4) { /* * And do a Getattr op. */ NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETATTR); NFSGETATTR_ATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); } error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); error = nfscl_wcc_data(nd, vp, nap, attrflagp, NULL, stuff); if (!error && !nd->nd_repstat) { NFSM_DISSECT(tl, u_int32_t *, NFSX_VERF); NFSLOCKMNT(nmp); if (NFSBCMP(nmp->nm_verf, tl, NFSX_VERF)) { NFSBCOPY(tl, nmp->nm_verf, NFSX_VERF); nd->nd_repstat = NFSERR_STALEWRITEVERF; } NFSUNLOCKMNT(nmp); if (nd->nd_flag & ND_NFSV4) error = nfscl_postop_attr(nd, nap, attrflagp, stuff); } nfsmout: if (!error && nd->nd_repstat) error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * NFS byte range lock rpc. * (Mostly just calls one of the three lower level RPC routines.) */ APPLESTATIC int nfsrpc_advlock(vnode_t vp, off_t size, int op, struct flock *fl, int reclaim, struct ucred *cred, NFSPROC_T *p, void *id, int flags) { struct nfscllockowner *lp; struct nfsclclient *clp; struct nfsfh *nfhp; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); u_int64_t off, len; off_t start, end; u_int32_t clidrev = 0; int error = 0, newone = 0, expireret = 0, retrycnt, donelocally; int callcnt, dorpc; /* * Convert the flock structure into a start and end and do POSIX * bounds checking. */ switch (fl->l_whence) { case SEEK_SET: case SEEK_CUR: /* * Caller is responsible for adding any necessary offset * when SEEK_CUR is used. */ start = fl->l_start; off = fl->l_start; break; case SEEK_END: start = size + fl->l_start; off = size + fl->l_start; break; default: return (EINVAL); } if (start < 0) return (EINVAL); if (fl->l_len != 0) { end = start + fl->l_len - 1; if (end < start) return (EINVAL); } len = fl->l_len; if (len == 0) len = NFS64BITSSET; retrycnt = 0; do { nd->nd_repstat = 0; if (op == F_GETLK) { error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp); if (error) return (error); error = nfscl_lockt(vp, clp, off, len, fl, p, id, flags); if (!error) { clidrev = clp->nfsc_clientidrev; error = nfsrpc_lockt(nd, vp, clp, off, len, fl, cred, p, id, flags); } else if (error == -1) { error = 0; } nfscl_clientrelease(clp); } else if (op == F_UNLCK && fl->l_type == F_UNLCK) { /* * We must loop around for all lockowner cases. */ callcnt = 0; error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp); if (error) return (error); do { error = nfscl_relbytelock(vp, off, len, cred, p, callcnt, clp, id, flags, &lp, &dorpc); /* * If it returns a NULL lp, we're done. */ if (lp == NULL) { if (callcnt == 0) nfscl_clientrelease(clp); else nfscl_releasealllocks(clp, vp, p, id, flags); return (error); } if (nmp->nm_clp != NULL) clidrev = nmp->nm_clp->nfsc_clientidrev; else clidrev = 0; /* * If the server doesn't support Posix lock semantics, * only allow locks on the entire file, since it won't * handle overlapping byte ranges. * There might still be a problem when a lock * upgrade/downgrade (read<->write) occurs, since the * server "might" expect an unlock first? */ if (dorpc && (lp->nfsl_open->nfso_posixlock || (off == 0 && len == NFS64BITSSET))) { /* * Since the lock records will go away, we must * wait for grace and delay here. */ do { error = nfsrpc_locku(nd, nmp, lp, off, len, NFSV4LOCKT_READ, cred, p, 0); if ((nd->nd_repstat == NFSERR_GRACE || nd->nd_repstat == NFSERR_DELAY) && error == 0) (void) nfs_catnap(PZERO, (int)nd->nd_repstat, "nfs_advlock"); } while ((nd->nd_repstat == NFSERR_GRACE || nd->nd_repstat == NFSERR_DELAY) && error == 0); } callcnt++; } while (error == 0 && nd->nd_repstat == 0); nfscl_releasealllocks(clp, vp, p, id, flags); } else if (op == F_SETLK) { error = nfscl_getbytelock(vp, off, len, fl->l_type, cred, p, NULL, 0, id, flags, NULL, NULL, &lp, &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; if (!lp->nfsl_open->nfso_posixlock && (off != 0 || len != NFS64BITSSET)) { error = EINVAL; } else { error = nfsrpc_lock(nd, nmp, vp, nfhp->nfh_fh, nfhp->nfh_len, lp, newone, reclaim, off, len, fl->l_type, cred, p, 0); } if (!error) error = nd->nd_repstat; nfscl_lockrelease(lp, error, newone); } else { error = EINVAL; } if (!error) error = nd->nd_repstat; if (error == NFSERR_GRACE || error == NFSERR_STALESTATEID || error == NFSERR_STALEDONTRECOVER || error == NFSERR_STALECLIENTID || error == NFSERR_DELAY || error == NFSERR_BADSESSION) { (void) nfs_catnap(PZERO, error, "nfs_advlock"); } else if ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && clidrev != 0) { expireret = nfscl_hasexpired(nmp->nm_clp, clidrev, p); retrycnt++; } } while (error == NFSERR_GRACE || error == NFSERR_STALECLIENTID || error == NFSERR_DELAY || error == NFSERR_STALEDONTRECOVER || error == NFSERR_STALESTATEID || error == NFSERR_BADSESSION || ((error == NFSERR_EXPIRED || error == NFSERR_BADSTATEID) && expireret == 0 && clidrev != 0 && retrycnt < 4)); if (error && retrycnt >= 4) error = EIO; return (error); } /* * The lower level routine for the LockT case. */ APPLESTATIC int nfsrpc_lockt(struct nfsrv_descript *nd, vnode_t vp, struct nfsclclient *clp, u_int64_t off, u_int64_t len, struct flock *fl, struct ucred *cred, NFSPROC_T *p, void *id, int flags) { u_int32_t *tl; int error, type, size; uint8_t own[NFSV4CL_LOCKNAMELEN + NFSX_V4FHMAX]; struct nfsnode *np; struct nfsmount *nmp; struct nfsclsession *tsep; nmp = VFSTONFS(vp->v_mount); NFSCL_REQSTART(nd, NFSPROC_LOCKT, vp); NFSM_BUILD(tl, u_int32_t *, 7 * NFSX_UNSIGNED); if (fl->l_type == F_RDLCK) *tl++ = txdr_unsigned(NFSV4LOCKT_READ); else *tl++ = txdr_unsigned(NFSV4LOCKT_WRITE); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; nfscl_filllockowner(id, own, flags); np = VTONFS(vp); NFSBCOPY(np->n_fhp->nfh_fh, &own[NFSV4CL_LOCKNAMELEN], np->n_fhp->nfh_len); (void)nfsm_strtom(nd, own, NFSV4CL_LOCKNAMELEN + np->n_fhp->nfh_len); error = nfscl_request(nd, vp, p, cred, NULL); if (error) return (error); if (nd->nd_repstat == 0) { fl->l_type = F_UNLCK; } else if (nd->nd_repstat == NFSERR_DENIED) { nd->nd_repstat = 0; fl->l_whence = SEEK_SET; NFSM_DISSECT(tl, u_int32_t *, 8 * NFSX_UNSIGNED); fl->l_start = fxdr_hyper(tl); tl += 2; len = fxdr_hyper(tl); tl += 2; if (len == NFS64BITSSET) fl->l_len = 0; else fl->l_len = len; type = fxdr_unsigned(int, *tl++); if (type == NFSV4LOCKT_WRITE) fl->l_type = F_WRLCK; else fl->l_type = F_RDLCK; /* * XXX For now, I have no idea what to do with the * conflicting lock_owner, so I'll just set the pid == 0 * and skip over the lock_owner. */ fl->l_pid = (pid_t)0; tl += 2; size = fxdr_unsigned(int, *tl); if (size < 0 || size > NFSV4_OPAQUELIMIT) error = EBADRPC; if (!error) error = nfsm_advance(nd, NFSM_RNDUP(size), -1); } else if (nd->nd_repstat == NFSERR_STALECLIENTID) nfscl_initiate_recovery(clp); nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Lower level function that performs the LockU RPC. */ static int nfsrpc_locku(struct nfsrv_descript *nd, struct nfsmount *nmp, struct nfscllockowner *lp, u_int64_t off, u_int64_t len, u_int32_t type, struct ucred *cred, NFSPROC_T *p, int syscred) { u_int32_t *tl; int error; nfscl_reqstart(nd, NFSPROC_LOCKU, nmp, lp->nfsl_open->nfso_fh, lp->nfsl_open->nfso_fhlen, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID + 6 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(type); *tl = txdr_unsigned(lp->nfsl_seqid); if (nfstest_outofseq && (arc4random() % nfstest_outofseq) == 0) *tl = txdr_unsigned(lp->nfsl_seqid + 1); tl++; if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = lp->nfsl_stateid.seqid; *tl++ = lp->nfsl_stateid.other[0]; *tl++ = lp->nfsl_stateid.other[1]; *tl++ = lp->nfsl_stateid.other[2]; txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); if (syscred) nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); NFSCL_INCRSEQID(lp->nfsl_seqid, nd); if (error) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID); lp->nfsl_stateid.seqid = *tl++; lp->nfsl_stateid.other[0] = *tl++; lp->nfsl_stateid.other[1] = *tl++; lp->nfsl_stateid.other[2] = *tl; } else if (nd->nd_repstat == NFSERR_STALESTATEID) nfscl_initiate_recovery(lp->nfsl_open->nfso_own->nfsow_clp); nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * The actual Lock RPC. */ APPLESTATIC int nfsrpc_lock(struct nfsrv_descript *nd, struct nfsmount *nmp, vnode_t vp, u_int8_t *nfhp, int fhlen, struct nfscllockowner *lp, int newone, int reclaim, u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p, int syscred) { u_int32_t *tl; int error, size; uint8_t own[NFSV4CL_LOCKNAMELEN + NFSX_V4FHMAX]; struct nfsclsession *tsep; nfscl_reqstart(nd, NFSPROC_LOCK, nmp, nfhp, fhlen, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, 7 * NFSX_UNSIGNED); if (type == F_RDLCK) *tl++ = txdr_unsigned(NFSV4LOCKT_READ); else *tl++ = txdr_unsigned(NFSV4LOCKT_WRITE); *tl++ = txdr_unsigned(reclaim); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; if (newone) { *tl = newnfs_true; NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID + 2 * NFSX_UNSIGNED + NFSX_HYPER); *tl++ = txdr_unsigned(lp->nfsl_open->nfso_own->nfsow_seqid); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = lp->nfsl_open->nfso_stateid.seqid; *tl++ = lp->nfsl_open->nfso_stateid.other[0]; *tl++ = lp->nfsl_open->nfso_stateid.other[1]; *tl++ = lp->nfsl_open->nfso_stateid.other[2]; *tl++ = txdr_unsigned(lp->nfsl_seqid); tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; NFSBCOPY(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN); NFSBCOPY(nfhp, &own[NFSV4CL_LOCKNAMELEN], fhlen); (void)nfsm_strtom(nd, own, NFSV4CL_LOCKNAMELEN + fhlen); } else { *tl = newnfs_false; NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID + NFSX_UNSIGNED); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = lp->nfsl_stateid.seqid; *tl++ = lp->nfsl_stateid.other[0]; *tl++ = lp->nfsl_stateid.other[1]; *tl++ = lp->nfsl_stateid.other[2]; *tl = txdr_unsigned(lp->nfsl_seqid); if (nfstest_outofseq && (arc4random() % nfstest_outofseq) == 0) *tl = txdr_unsigned(lp->nfsl_seqid + 1); } if (syscred) nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); if (newone) NFSCL_INCRSEQID(lp->nfsl_open->nfso_own->nfsow_seqid, nd); NFSCL_INCRSEQID(lp->nfsl_seqid, nd); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID); lp->nfsl_stateid.seqid = *tl++; lp->nfsl_stateid.other[0] = *tl++; lp->nfsl_stateid.other[1] = *tl++; lp->nfsl_stateid.other[2] = *tl; } else if (nd->nd_repstat == NFSERR_DENIED) { NFSM_DISSECT(tl, u_int32_t *, 8 * NFSX_UNSIGNED); size = fxdr_unsigned(int, *(tl + 7)); if (size < 0 || size > NFSV4_OPAQUELIMIT) error = EBADRPC; if (!error) error = nfsm_advance(nd, NFSM_RNDUP(size), -1); } else if (nd->nd_repstat == NFSERR_STALESTATEID) nfscl_initiate_recovery(lp->nfsl_open->nfso_own->nfsow_clp); nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs statfs rpc * (always called with the vp for the mount point) */ APPLESTATIC int nfsrpc_statfs(vnode_t vp, struct nfsstatfs *sbp, struct nfsfsinfo *fsp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { u_int32_t *tl = NULL; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp; nfsattrbit_t attrbits; int error; *attrflagp = 0; nmp = VFSTONFS(vnode_mount(vp)); if (NFSHASNFSV4(nmp)) { /* * For V4, you actually do a getattr. */ NFSCL_REQSTART(nd, NFSPROC_GETATTR, vp); NFSSTATFS_GETATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); nd->nd_flag |= ND_USEGSSNAME; error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_repstat == 0) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL, sbp, fsp, NULL, 0, NULL, NULL, NULL, p, cred); if (!error) { nmp->nm_fsid[0] = nap->na_filesid[0]; nmp->nm_fsid[1] = nap->na_filesid[1]; NFSSETHASSETFSID(nmp); *attrflagp = 1; } } else { error = nd->nd_repstat; } if (error) goto nfsmout; } else { NFSCL_REQSTART(nd, NFSPROC_FSSTAT, vp); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_flag & ND_NFSV3) { error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (error) goto nfsmout; } if (nd->nd_repstat) { error = nd->nd_repstat; goto nfsmout; } NFSM_DISSECT(tl, u_int32_t *, NFSX_STATFS(nd->nd_flag & ND_NFSV3)); } if (NFSHASNFSV3(nmp)) { sbp->sf_tbytes = fxdr_hyper(tl); tl += 2; sbp->sf_fbytes = fxdr_hyper(tl); tl += 2; sbp->sf_abytes = fxdr_hyper(tl); tl += 2; sbp->sf_tfiles = fxdr_hyper(tl); tl += 2; sbp->sf_ffiles = fxdr_hyper(tl); tl += 2; sbp->sf_afiles = fxdr_hyper(tl); tl += 2; sbp->sf_invarsec = fxdr_unsigned(u_int32_t, *tl); } else if (NFSHASNFSV4(nmp) == 0) { sbp->sf_tsize = fxdr_unsigned(u_int32_t, *tl++); sbp->sf_bsize = fxdr_unsigned(u_int32_t, *tl++); sbp->sf_blocks = fxdr_unsigned(u_int32_t, *tl++); sbp->sf_bfree = fxdr_unsigned(u_int32_t, *tl++); sbp->sf_bavail = fxdr_unsigned(u_int32_t, *tl); } nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs pathconf rpc */ APPLESTATIC int nfsrpc_pathconf(vnode_t vp, struct nfsv3_pathconf *pc, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp; u_int32_t *tl; nfsattrbit_t attrbits; int error; *attrflagp = 0; nmp = VFSTONFS(vnode_mount(vp)); if (NFSHASNFSV4(nmp)) { /* * For V4, you actually do a getattr. */ NFSCL_REQSTART(nd, NFSPROC_GETATTR, vp); NFSPATHCONF_GETATTRBIT(&attrbits); (void) nfsrv_putattrbit(nd, &attrbits); nd->nd_flag |= ND_USEGSSNAME; error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (nd->nd_repstat == 0) { error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, pc, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, p, cred); if (!error) *attrflagp = 1; } else { error = nd->nd_repstat; } } else { NFSCL_REQSTART(nd, NFSPROC_PATHCONF, vp); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (nd->nd_repstat && !error) error = nd->nd_repstat; if (!error) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V3PATHCONF); pc->pc_linkmax = fxdr_unsigned(u_int32_t, *tl++); pc->pc_namemax = fxdr_unsigned(u_int32_t, *tl++); pc->pc_notrunc = fxdr_unsigned(u_int32_t, *tl++); pc->pc_chownrestricted = fxdr_unsigned(u_int32_t, *tl++); pc->pc_caseinsensitive = fxdr_unsigned(u_int32_t, *tl++); pc->pc_casepreserving = fxdr_unsigned(u_int32_t, *tl); } } nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * nfs version 3 fsinfo rpc call */ APPLESTATIC int nfsrpc_fsinfo(vnode_t vp, struct nfsfsinfo *fsp, struct ucred *cred, NFSPROC_T *p, struct nfsvattr *nap, int *attrflagp, void *stuff) { u_int32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; int error; *attrflagp = 0; NFSCL_REQSTART(nd, NFSPROC_FSINFO, vp); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); error = nfscl_postop_attr(nd, nap, attrflagp, stuff); if (nd->nd_repstat && !error) error = nd->nd_repstat; if (!error) { NFSM_DISSECT(tl, u_int32_t *, NFSX_V3FSINFO); fsp->fs_rtmax = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_rtpref = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_rtmult = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_wtmax = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_wtpref = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_wtmult = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_dtpref = fxdr_unsigned(u_int32_t, *tl++); fsp->fs_maxfilesize = fxdr_hyper(tl); tl += 2; fxdr_nfsv3time(tl, &fsp->fs_timedelta); tl += 2; fsp->fs_properties = fxdr_unsigned(u_int32_t, *tl); } nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * This function performs the Renew RPC. */ APPLESTATIC int nfsrpc_renew(struct nfsclclient *clp, struct nfsclds *dsp, struct ucred *cred, NFSPROC_T *p) { u_int32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; struct nfsmount *nmp; int error; struct nfssockreq *nrp; struct nfsclsession *tsep; nmp = clp->nfsc_nmp; if (nmp == NULL) return (0); if (dsp == NULL) nfscl_reqstart(nd, NFSPROC_RENEW, nmp, NULL, 0, NULL, NULL); else nfscl_reqstart(nd, NFSPROC_RENEW, nmp, NULL, 0, NULL, &dsp->nfsclds_sess); if (!NFSHASNFSV4N(nmp)) { /* NFSv4.1 just uses a Sequence Op and not a Renew. */ NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; } nrp = NULL; if (dsp != NULL) nrp = dsp->nfsclds_sockp; if (nrp == NULL) /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; nd->nd_flag |= ND_USEGSSNAME; if (dsp == NULL) error = newnfs_request(nd, nmp, NULL, nrp, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); else error = newnfs_request(nd, nmp, NULL, nrp, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); if (error) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * This function performs the Releaselockowner RPC. */ APPLESTATIC int nfsrpc_rellockown(struct nfsmount *nmp, struct nfscllockowner *lp, uint8_t *fh, int fhlen, struct ucred *cred, NFSPROC_T *p) { struct nfsrv_descript nfsd, *nd = &nfsd; u_int32_t *tl; int error; uint8_t own[NFSV4CL_LOCKNAMELEN + NFSX_V4FHMAX]; struct nfsclsession *tsep; if (NFSHASNFSV4N(nmp)) { /* For NFSv4.1, do a FreeStateID. */ nfscl_reqstart(nd, NFSPROC_FREESTATEID, nmp, NULL, 0, NULL, NULL); nfsm_stateidtom(nd, &lp->nfsl_stateid, NFSSTATEID_PUTSTATEID); } else { nfscl_reqstart(nd, NFSPROC_RELEASELCKOWN, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED); tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; NFSBCOPY(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN); NFSBCOPY(fh, &own[NFSV4CL_LOCKNAMELEN], fhlen); (void)nfsm_strtom(nd, own, NFSV4CL_LOCKNAMELEN + fhlen); } nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * This function performs the Compound to get the mount pt FH. */ APPLESTATIC int nfsrpc_getdirpath(struct nfsmount *nmp, u_char *dirpath, struct ucred *cred, NFSPROC_T *p) { u_int32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; u_char *cp, *cp2; int error, cnt, len, setnil; u_int32_t *opcntp; nfscl_reqstart(nd, NFSPROC_PUTROOTFH, nmp, NULL, 0, &opcntp, NULL); cp = dirpath; cnt = 0; do { setnil = 0; while (*cp == '/') cp++; cp2 = cp; while (*cp2 != '\0' && *cp2 != '/') cp2++; if (*cp2 == '/') { setnil = 1; *cp2 = '\0'; } if (cp2 != cp) { NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_LOOKUP); nfsm_strtom(nd, cp, strlen(cp)); cnt++; } if (setnil) *cp2++ = '/'; cp = cp2; } while (*cp != '\0'); if (NFSHASNFSV4N(nmp)) /* Has a Sequence Op done by nfscl_reqstart(). */ *opcntp = txdr_unsigned(3 + cnt); else *opcntp = txdr_unsigned(2 + cnt); NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(NFSV4OP_GETFH); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, (3 + 2 * cnt) * NFSX_UNSIGNED); tl += (2 + 2 * cnt); if ((len = fxdr_unsigned(int, *tl)) <= 0 || len > NFSX_FHMAX) { nd->nd_repstat = NFSERR_BADXDR; } else { nd->nd_repstat = nfsrv_mtostr(nd, nmp->nm_fh, len); if (nd->nd_repstat == 0) nmp->nm_fhsize = len; } } error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * This function performs the Delegreturn RPC. */ APPLESTATIC int nfsrpc_delegreturn(struct nfscldeleg *dp, struct ucred *cred, struct nfsmount *nmp, NFSPROC_T *p, int syscred) { u_int32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; int error; nfscl_reqstart(nd, NFSPROC_DELEGRETURN, nmp, dp->nfsdl_fh, dp->nfsdl_fhlen, NULL, NULL); NFSM_BUILD(tl, u_int32_t *, NFSX_STATEID); if (NFSHASNFSV4N(nmp)) *tl++ = 0; else *tl++ = dp->nfsdl_stateid.seqid; *tl++ = dp->nfsdl_stateid.other[0]; *tl++ = dp->nfsdl_stateid.other[1]; *tl = dp->nfsdl_stateid.other[2]; if (syscred) nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * nfs getacl call. */ APPLESTATIC int nfsrpc_getacl(vnode_t vp, struct ucred *cred, NFSPROC_T *p, struct acl *aclp, void *stuff) { struct nfsrv_descript nfsd, *nd = &nfsd; int error; nfsattrbit_t attrbits; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); if (nfsrv_useacl == 0 || !NFSHASNFSV4(nmp)) return (EOPNOTSUPP); NFSCL_REQSTART(nd, NFSPROC_GETACL, vp); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); (void) nfsrv_putattrbit(nd, &attrbits); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); if (!nd->nd_repstat) error = nfsv4_loadattr(nd, vp, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, aclp, 0, NULL, NULL, NULL, p, cred); else error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * nfs setacl call. */ APPLESTATIC int nfsrpc_setacl(vnode_t vp, struct ucred *cred, NFSPROC_T *p, struct acl *aclp, void *stuff) { int error; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); if (nfsrv_useacl == 0 || !NFSHASNFSV4(nmp)) return (EOPNOTSUPP); error = nfsrpc_setattr(vp, NULL, aclp, cred, p, NULL, NULL, stuff); return (error); } /* * nfs setacl call. */ static int nfsrpc_setaclrpc(vnode_t vp, struct ucred *cred, NFSPROC_T *p, struct acl *aclp, nfsv4stateid_t *stateidp, void *stuff) { struct nfsrv_descript nfsd, *nd = &nfsd; int error; nfsattrbit_t attrbits; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); if (!NFSHASNFSV4(nmp)) return (EOPNOTSUPP); NFSCL_REQSTART(nd, NFSPROC_SETACL, vp); nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSTATEID); NFSZERO_ATTRBIT(&attrbits); NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL); (void) nfsv4_fillattr(nd, vnode_mount(vp), vp, aclp, NULL, NULL, 0, &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0); error = nfscl_request(nd, vp, p, cred, stuff); if (error) return (error); /* Don't care about the pre/postop attributes */ mbuf_freem(nd->nd_mrep); return (nd->nd_repstat); } /* * Do the NFSv4.1 Exchange ID. */ int nfsrpc_exchangeid(struct nfsmount *nmp, struct nfsclclient *clp, struct nfssockreq *nrp, uint32_t exchflags, struct nfsclds **dspp, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl, v41flags; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; struct nfsclds *dsp; struct timespec verstime; int error, len; *dspp = NULL; nfscl_reqstart(nd, NFSPROC_EXCHANGEID, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, uint32_t *, 2 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(nfsboottime.tv_sec); /* Client owner */ *tl = txdr_unsigned(clp->nfsc_rev); (void) nfsm_strtom(nd, clp->nfsc_id, clp->nfsc_idlen); NFSM_BUILD(tl, uint32_t *, 3 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(exchflags); *tl++ = txdr_unsigned(NFSV4EXCH_SP4NONE); /* Set the implementation id4 */ *tl = txdr_unsigned(1); (void) nfsm_strtom(nd, "freebsd.org", strlen("freebsd.org")); (void) nfsm_strtom(nd, version, strlen(version)); NFSM_BUILD(tl, uint32_t *, NFSX_V4TIME); verstime.tv_sec = 1293840000; /* Jan 1, 2011 */ verstime.tv_nsec = 0; txdr_nfsv4time(&verstime, tl); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, nrp, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); NFSCL_DEBUG(1, "exchangeid err=%d reps=%d\n", error, (int)nd->nd_repstat); if (error != 0) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 6 * NFSX_UNSIGNED + NFSX_HYPER); len = fxdr_unsigned(int, *(tl + 7)); if (len < 0 || len > NFSV4_OPAQUELIMIT) { error = NFSERR_BADXDR; goto nfsmout; } dsp = malloc(sizeof(struct nfsclds) + len + 1, M_NFSCLDS, M_WAITOK | M_ZERO); dsp->nfsclds_expire = NFSD_MONOSEC + clp->nfsc_renew; dsp->nfsclds_servownlen = len; dsp->nfsclds_sess.nfsess_clientid.lval[0] = *tl++; dsp->nfsclds_sess.nfsess_clientid.lval[1] = *tl++; dsp->nfsclds_sess.nfsess_sequenceid = fxdr_unsigned(uint32_t, *tl++); v41flags = fxdr_unsigned(uint32_t, *tl); if ((v41flags & NFSV4EXCH_USEPNFSMDS) != 0 && NFSHASPNFSOPT(nmp)) { NFSCL_DEBUG(1, "set PNFS\n"); NFSLOCKMNT(nmp); nmp->nm_state |= NFSSTA_PNFS; NFSUNLOCKMNT(nmp); dsp->nfsclds_flags |= NFSCLDS_MDS; } if ((v41flags & NFSV4EXCH_USEPNFSDS) != 0) dsp->nfsclds_flags |= NFSCLDS_DS; if (len > 0) nd->nd_repstat = nfsrv_mtostr(nd, dsp->nfsclds_serverown, len); if (nd->nd_repstat == 0) { mtx_init(&dsp->nfsclds_mtx, "nfsds", NULL, MTX_DEF); mtx_init(&dsp->nfsclds_sess.nfsess_mtx, "nfssession", NULL, MTX_DEF); nfscl_initsessionslots(&dsp->nfsclds_sess); *dspp = dsp; } else free(dsp, M_NFSCLDS); } error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 Create Session. */ int nfsrpc_createsession(struct nfsmount *nmp, struct nfsclsession *sep, struct nfssockreq *nrp, uint32_t sequenceid, int mds, struct ucred *cred, NFSPROC_T *p) { - uint32_t crflags, *tl; + uint32_t crflags, maxval, *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; int error, irdcnt; nfscl_reqstart(nd, NFSPROC_CREATESESSION, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED); *tl++ = sep->nfsess_clientid.lval[0]; *tl++ = sep->nfsess_clientid.lval[1]; *tl++ = txdr_unsigned(sequenceid); crflags = (NFSMNT_RDONLY(nmp->nm_mountp) ? 0 : NFSV4CRSESS_PERSIST); if (nfscl_enablecallb != 0 && nfs_numnfscbd > 0 && mds != 0) crflags |= NFSV4CRSESS_CONNBACKCHAN; *tl = txdr_unsigned(crflags); /* Fill in fore channel attributes. */ NFSM_BUILD(tl, uint32_t *, 7 * NFSX_UNSIGNED); *tl++ = 0; /* Header pad size */ - *tl++ = txdr_unsigned(100000); /* Max request size */ - *tl++ = txdr_unsigned(100000); /* Max response size */ + *tl++ = txdr_unsigned(nmp->nm_wsize + NFS_MAXXDR);/* Max request size */ + *tl++ = txdr_unsigned(nmp->nm_rsize + NFS_MAXXDR);/* Max reply size */ *tl++ = txdr_unsigned(4096); /* Max response size cached */ *tl++ = txdr_unsigned(20); /* Max operations */ *tl++ = txdr_unsigned(64); /* Max slots */ *tl = 0; /* No rdma ird */ /* Fill in back channel attributes. */ NFSM_BUILD(tl, uint32_t *, 7 * NFSX_UNSIGNED); *tl++ = 0; /* Header pad size */ *tl++ = txdr_unsigned(10000); /* Max request size */ *tl++ = txdr_unsigned(10000); /* Max response size */ *tl++ = txdr_unsigned(4096); /* Max response size cached */ *tl++ = txdr_unsigned(4); /* Max operations */ *tl++ = txdr_unsigned(NFSV4_CBSLOTS); /* Max slots */ *tl = 0; /* No rdma ird */ NFSM_BUILD(tl, uint32_t *, 8 * NFSX_UNSIGNED); *tl++ = txdr_unsigned(NFS_CALLBCKPROG); /* Call back prog # */ /* Allow AUTH_SYS callbacks as uid, gid == 0. */ *tl++ = txdr_unsigned(1); /* Auth_sys only */ *tl++ = txdr_unsigned(AUTH_SYS); /* AUTH_SYS type */ *tl++ = txdr_unsigned(nfsboottime.tv_sec); /* time stamp */ *tl++ = 0; /* Null machine name */ *tl++ = 0; /* Uid == 0 */ *tl++ = 0; /* Gid == 0 */ *tl = 0; /* No additional gids */ nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, nrp, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID + 2 * NFSX_UNSIGNED); bcopy(tl, sep->nfsess_sessionid, NFSX_V4SESSIONID); tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; sep->nfsess_sequenceid = fxdr_unsigned(uint32_t, *tl++); crflags = fxdr_unsigned(uint32_t, *tl); if ((crflags & NFSV4CRSESS_PERSIST) != 0 && mds != 0) { NFSLOCKMNT(nmp); nmp->nm_state |= NFSSTA_SESSPERSIST; NFSUNLOCKMNT(nmp); } /* Get the fore channel slot count. */ NFSM_DISSECT(tl, uint32_t *, 7 * NFSX_UNSIGNED); - tl += 3; /* Skip the other counts. */ + tl++; /* Skip the header pad size. */ + + /* Make sure nm_wsize is small enough. */ + maxval = fxdr_unsigned(uint32_t, *tl++); + while (maxval < nmp->nm_wsize + NFS_MAXXDR) { + if (nmp->nm_wsize > 8096) + nmp->nm_wsize /= 2; + else + break; + } + + /* Make sure nm_rsize is small enough. */ + maxval = fxdr_unsigned(uint32_t, *tl++); + while (maxval < nmp->nm_rsize + NFS_MAXXDR) { + if (nmp->nm_rsize > 8096) + nmp->nm_rsize /= 2; + else + break; + } + sep->nfsess_maxcache = fxdr_unsigned(int, *tl++); tl++; sep->nfsess_foreslots = fxdr_unsigned(uint16_t, *tl++); NFSCL_DEBUG(4, "fore slots=%d\n", (int)sep->nfsess_foreslots); irdcnt = fxdr_unsigned(int, *tl); if (irdcnt > 0) NFSM_DISSECT(tl, uint32_t *, irdcnt * NFSX_UNSIGNED); /* and the back channel slot count. */ NFSM_DISSECT(tl, uint32_t *, 7 * NFSX_UNSIGNED); tl += 5; sep->nfsess_backslots = fxdr_unsigned(uint16_t, *tl); NFSCL_DEBUG(4, "back slots=%d\n", (int)sep->nfsess_backslots); } error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 Destroy Session. */ int nfsrpc_destroysession(struct nfsmount *nmp, struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; int error; struct nfsclsession *tsep; nfscl_reqstart(nd, NFSPROC_DESTROYSESSION, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID); tsep = nfsmnt_mdssession(nmp); bcopy(tsep->nfsess_sessionid, tl, NFSX_V4SESSIONID); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 Destroy Client. */ int nfsrpc_destroyclient(struct nfsmount *nmp, struct nfsclclient *clp, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; int error; struct nfsclsession *tsep; nfscl_reqstart(nd, NFSPROC_DESTROYCLIENT, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, uint32_t *, 2 * NFSX_UNSIGNED); tsep = nfsmnt_mdssession(nmp); *tl++ = tsep->nfsess_clientid.lval[0]; *tl = tsep->nfsess_clientid.lval[1]; nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 LayoutGet. */ int nfsrpc_layoutget(struct nfsmount *nmp, uint8_t *fhp, int fhlen, int iomode, uint64_t offset, uint64_t len, uint64_t minlen, int layoutlen, nfsv4stateid_t *stateidp, int *retonclosep, struct nfsclflayouthead *flhp, struct ucred *cred, NFSPROC_T *p, void *stuff) { uint32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsfh *nfhp; struct nfsclflayout *flp, *prevflp, *tflp; int cnt, error, gotiomode, fhcnt, nfhlen, i, j; uint8_t *cp; uint64_t retlen; flp = NULL; gotiomode = -1; nfscl_reqstart(nd, NFSPROC_LAYOUTGET, nmp, fhp, fhlen, NULL, NULL); NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED + 3 * NFSX_HYPER + NFSX_STATEID); *tl++ = newnfs_false; /* Don't signal availability. */ *tl++ = txdr_unsigned(NFSLAYOUT_NFSV4_1_FILES); *tl++ = txdr_unsigned(iomode); txdr_hyper(offset, tl); tl += 2; txdr_hyper(len, tl); tl += 2; txdr_hyper(minlen, tl); tl += 2; *tl++ = txdr_unsigned(stateidp->seqid); NFSCL_DEBUG(4, "layget seq=%d\n", (int)stateidp->seqid); *tl++ = stateidp->other[0]; *tl++ = stateidp->other[1]; *tl++ = stateidp->other[2]; *tl = txdr_unsigned(layoutlen); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_STATEID); if (*tl++ != 0) *retonclosep = 1; else *retonclosep = 0; stateidp->seqid = fxdr_unsigned(uint32_t, *tl++); NFSCL_DEBUG(4, "retoncls=%d stseq=%d\n", *retonclosep, (int)stateidp->seqid); stateidp->other[0] = *tl++; stateidp->other[1] = *tl++; stateidp->other[2] = *tl++; cnt = fxdr_unsigned(int, *tl); NFSCL_DEBUG(4, "layg cnt=%d\n", cnt); if (cnt <= 0 || cnt > 10000) { /* Don't accept more than 10000 layouts in reply. */ error = NFSERR_BADXDR; goto nfsmout; } for (i = 0; i < cnt; i++) { /* Dissect all the way to the file handle cnt. */ NFSM_DISSECT(tl, uint32_t *, 3 * NFSX_HYPER + 6 * NFSX_UNSIGNED + NFSX_V4DEVICEID); fhcnt = fxdr_unsigned(int, *(tl + 11 + NFSX_V4DEVICEID / NFSX_UNSIGNED)); NFSCL_DEBUG(4, "fhcnt=%d\n", fhcnt); if (fhcnt < 0 || fhcnt > 100) { /* Don't accept more than 100 file handles. */ error = NFSERR_BADXDR; goto nfsmout; } if (fhcnt > 1) flp = malloc(sizeof(*flp) + (fhcnt - 1) * sizeof(struct nfsfh *), M_NFSFLAYOUT, M_WAITOK); else flp = malloc(sizeof(*flp), M_NFSFLAYOUT, M_WAITOK); flp->nfsfl_flags = 0; flp->nfsfl_fhcnt = 0; flp->nfsfl_devp = NULL; flp->nfsfl_off = fxdr_hyper(tl); tl += 2; retlen = fxdr_hyper(tl); tl += 2; if (flp->nfsfl_off + retlen < flp->nfsfl_off) flp->nfsfl_end = UINT64_MAX - flp->nfsfl_off; else flp->nfsfl_end = flp->nfsfl_off + retlen; flp->nfsfl_iomode = fxdr_unsigned(int, *tl++); if (gotiomode == -1) gotiomode = flp->nfsfl_iomode; NFSCL_DEBUG(4, "layg reqiom=%d retiom=%d\n", iomode, (int)flp->nfsfl_iomode); if (fxdr_unsigned(int, *tl++) != NFSLAYOUT_NFSV4_1_FILES) { printf("NFSv4.1: got non-files layout\n"); error = NFSERR_BADXDR; goto nfsmout; } NFSBCOPY(++tl, flp->nfsfl_dev, NFSX_V4DEVICEID); tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); flp->nfsfl_util = fxdr_unsigned(uint32_t, *tl++); NFSCL_DEBUG(4, "flutil=0x%x\n", flp->nfsfl_util); flp->nfsfl_stripe1 = fxdr_unsigned(uint32_t, *tl++); flp->nfsfl_patoff = fxdr_hyper(tl); tl += 2; if (fxdr_unsigned(int, *tl) != fhcnt) { printf("EEK! bad fhcnt\n"); error = NFSERR_BADXDR; goto nfsmout; } for (j = 0; j < fhcnt; j++) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); nfhlen = fxdr_unsigned(int, *tl); if (nfhlen <= 0 || nfhlen > NFSX_V4FHMAX) { error = NFSERR_BADXDR; goto nfsmout; } nfhp = malloc(sizeof(*nfhp) + nfhlen - 1, M_NFSFH, M_WAITOK); flp->nfsfl_fh[j] = nfhp; flp->nfsfl_fhcnt++; nfhp->nfh_len = nfhlen; NFSM_DISSECT(cp, uint8_t *, NFSM_RNDUP(nfhlen)); NFSBCOPY(cp, nfhp->nfh_fh, nfhlen); } if (flp->nfsfl_iomode == gotiomode) { /* Keep the list in increasing offset order. */ tflp = LIST_FIRST(flhp); prevflp = NULL; while (tflp != NULL && tflp->nfsfl_off < flp->nfsfl_off) { prevflp = tflp; tflp = LIST_NEXT(tflp, nfsfl_list); } if (prevflp == NULL) LIST_INSERT_HEAD(flhp, flp, nfsfl_list); else LIST_INSERT_AFTER(prevflp, flp, nfsfl_list); } else { printf("nfscl_layoutget(): got wrong iomode\n"); nfscl_freeflayout(flp); } flp = NULL; } } if (nd->nd_repstat != 0 && error == 0) error = nd->nd_repstat; nfsmout: if (error != 0 && flp != NULL) nfscl_freeflayout(flp); mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 Get Device Info. */ int nfsrpc_getdeviceinfo(struct nfsmount *nmp, uint8_t *deviceid, int layouttype, uint32_t *notifybitsp, struct nfscldevinfo **ndip, struct ucred *cred, NFSPROC_T *p) { uint32_t cnt, *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; struct sockaddr_storage ss; struct nfsclds *dsp = NULL, **dspp; struct nfscldevinfo *ndi; int addrcnt, bitcnt, error, i, isudp, j, pos, safilled, stripecnt; uint8_t stripeindex; *ndip = NULL; ndi = NULL; nfscl_reqstart(nd, NFSPROC_GETDEVICEINFO, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, uint32_t *, NFSX_V4DEVICEID + 3 * NFSX_UNSIGNED); NFSBCOPY(deviceid, tl, NFSX_V4DEVICEID); tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); *tl++ = txdr_unsigned(layouttype); *tl++ = txdr_unsigned(100000); if (notifybitsp != NULL && *notifybitsp != 0) { *tl = txdr_unsigned(1); /* One word of bits. */ NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED); *tl = txdr_unsigned(*notifybitsp); } else *tl = txdr_unsigned(0); nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, 3 * NFSX_UNSIGNED); if (layouttype != fxdr_unsigned(int, *tl++)) printf("EEK! devinfo layout type not same!\n"); stripecnt = fxdr_unsigned(int, *++tl); NFSCL_DEBUG(4, "stripecnt=%d\n", stripecnt); if (stripecnt < 1 || stripecnt > 4096) { printf("NFS devinfo stripecnt %d: out of range\n", stripecnt); error = NFSERR_BADXDR; goto nfsmout; } NFSM_DISSECT(tl, uint32_t *, (stripecnt + 1) * NFSX_UNSIGNED); addrcnt = fxdr_unsigned(int, *(tl + stripecnt)); NFSCL_DEBUG(4, "addrcnt=%d\n", addrcnt); if (addrcnt < 1 || addrcnt > 128) { printf("NFS devinfo addrcnt %d: out of range\n", addrcnt); error = NFSERR_BADXDR; goto nfsmout; } /* * Now we know how many stripe indices and addresses, so * we can allocate the structure the correct size. */ i = (stripecnt * sizeof(uint8_t)) / sizeof(struct nfsclds *) + 1; NFSCL_DEBUG(4, "stripeindices=%d\n", i); ndi = malloc(sizeof(*ndi) + (addrcnt + i) * sizeof(struct nfsclds *), M_NFSDEVINFO, M_WAITOK | M_ZERO); NFSBCOPY(deviceid, ndi->nfsdi_deviceid, NFSX_V4DEVICEID); ndi->nfsdi_refcnt = 0; ndi->nfsdi_stripecnt = stripecnt; ndi->nfsdi_addrcnt = addrcnt; /* Fill in the stripe indices. */ for (i = 0; i < stripecnt; i++) { stripeindex = fxdr_unsigned(uint8_t, *tl++); NFSCL_DEBUG(4, "stripeind=%d\n", stripeindex); if (stripeindex >= addrcnt) { printf("NFS devinfo stripeindex %d: too big\n", (int)stripeindex); error = NFSERR_BADXDR; goto nfsmout; } nfsfldi_setstripeindex(ndi, i, stripeindex); } /* Now, dissect the server address(es). */ safilled = 0; for (i = 0; i < addrcnt; i++) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); cnt = fxdr_unsigned(uint32_t, *tl); if (cnt == 0) { printf("NFS devinfo 0 len addrlist\n"); error = NFSERR_BADXDR; goto nfsmout; } dspp = nfsfldi_addr(ndi, i); pos = arc4random() % cnt; /* Choose one. */ safilled = 0; for (j = 0; j < cnt; j++) { error = nfsv4_getipaddr(nd, &ss, &isudp); if (error != 0 && error != EPERM) { error = NFSERR_BADXDR; goto nfsmout; } if (error == 0 && isudp == 0) { /* * The algorithm is: * - use "pos" entry if it is of the * same af_family or none of them * is of the same af_family * else * - use the first one of the same * af_family. */ if ((safilled == 0 && ss.ss_family == nmp->nm_nam->sa_family) || (j == pos && (safilled == 0 || ss.ss_family == nmp->nm_nam->sa_family)) || (safilled == 1 && ss.ss_family == nmp->nm_nam->sa_family)) { error = nfsrpc_fillsa(nmp, &ss, &dsp, p); if (error == 0) { *dspp = dsp; if (ss.ss_family == nmp->nm_nam->sa_family) safilled = 2; else safilled = 1; } } } } if (safilled == 0) break; } /* And the notify bits. */ NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); if (safilled != 0) { bitcnt = fxdr_unsigned(int, *tl); if (bitcnt > 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); if (notifybitsp != NULL) *notifybitsp = fxdr_unsigned(uint32_t, *tl); } *ndip = ndi; } else error = EPERM; } if (nd->nd_repstat != 0) error = nd->nd_repstat; nfsmout: if (error != 0 && ndi != NULL) nfscl_freedevinfo(ndi); mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 LayoutCommit. */ int nfsrpc_layoutcommit(struct nfsmount *nmp, uint8_t *fh, int fhlen, int reclaim, uint64_t off, uint64_t len, uint64_t lastbyte, nfsv4stateid_t *stateidp, int layouttype, int layoutupdatecnt, uint8_t *layp, struct ucred *cred, NFSPROC_T *p, void *stuff) { uint32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; int error, outcnt, i; uint8_t *cp; nfscl_reqstart(nd, NFSPROC_LAYOUTCOMMIT, nmp, fh, fhlen, NULL, NULL); NFSM_BUILD(tl, uint32_t *, 5 * NFSX_UNSIGNED + 3 * NFSX_HYPER + NFSX_STATEID); txdr_hyper(off, tl); tl += 2; txdr_hyper(len, tl); tl += 2; if (reclaim != 0) *tl++ = newnfs_true; else *tl++ = newnfs_false; *tl++ = txdr_unsigned(stateidp->seqid); *tl++ = stateidp->other[0]; *tl++ = stateidp->other[1]; *tl++ = stateidp->other[2]; *tl++ = newnfs_true; if (lastbyte < off) lastbyte = off; else if (lastbyte >= (off + len)) lastbyte = off + len - 1; txdr_hyper(lastbyte, tl); tl += 2; *tl++ = newnfs_false; *tl++ = txdr_unsigned(layouttype); *tl = txdr_unsigned(layoutupdatecnt); if (layoutupdatecnt > 0) { KASSERT(layouttype != NFSLAYOUT_NFSV4_1_FILES, ("Must be nil for Files Layout")); outcnt = NFSM_RNDUP(layoutupdatecnt); NFSM_BUILD(cp, uint8_t *, outcnt); NFSBCOPY(layp, cp, layoutupdatecnt); cp += layoutupdatecnt; for (i = 0; i < (outcnt - layoutupdatecnt); i++) *cp++ = 0x0; } nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * Do the NFSv4.1 LayoutReturn. */ int nfsrpc_layoutreturn(struct nfsmount *nmp, uint8_t *fh, int fhlen, int reclaim, int layouttype, uint32_t iomode, int layoutreturn, uint64_t offset, uint64_t len, nfsv4stateid_t *stateidp, int layoutcnt, uint32_t *layp, struct ucred *cred, NFSPROC_T *p, void *stuff) { uint32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; int error, outcnt, i; uint8_t *cp; nfscl_reqstart(nd, NFSPROC_LAYOUTRETURN, nmp, fh, fhlen, NULL, NULL); NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED); if (reclaim != 0) *tl++ = newnfs_true; else *tl++ = newnfs_false; *tl++ = txdr_unsigned(layouttype); *tl++ = txdr_unsigned(iomode); *tl = txdr_unsigned(layoutreturn); if (layoutreturn == NFSLAYOUTRETURN_FILE) { NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_STATEID + NFSX_UNSIGNED); txdr_hyper(offset, tl); tl += 2; txdr_hyper(len, tl); tl += 2; NFSCL_DEBUG(4, "layoutret stseq=%d\n", (int)stateidp->seqid); *tl++ = txdr_unsigned(stateidp->seqid); *tl++ = stateidp->other[0]; *tl++ = stateidp->other[1]; *tl++ = stateidp->other[2]; *tl = txdr_unsigned(layoutcnt); if (layoutcnt > 0) { outcnt = NFSM_RNDUP(layoutcnt); NFSM_BUILD(cp, uint8_t *, outcnt); NFSBCOPY(layp, cp, layoutcnt); cp += layoutcnt; for (i = 0; i < (outcnt - layoutcnt); i++) *cp++ = 0x0; } } nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); if (*tl != 0) { NFSM_DISSECT(tl, uint32_t *, NFSX_STATEID); stateidp->seqid = fxdr_unsigned(uint32_t, *tl++); stateidp->other[0] = *tl++; stateidp->other[1] = *tl++; stateidp->other[2] = *tl; } } else error = nd->nd_repstat; nfsmout: mbuf_freem(nd->nd_mrep); return (error); } /* * Acquire a layout and devinfo, if possible. The caller must have acquired * a reference count on the nfsclclient structure before calling this. * Return the layout in lypp with a reference count on it, if successful. */ static int nfsrpc_getlayout(struct nfsmount *nmp, vnode_t vp, struct nfsfh *nfhp, int iomode, uint32_t *notifybitsp, nfsv4stateid_t *stateidp, uint64_t off, struct nfscllayout **lypp, struct ucred *cred, NFSPROC_T *p) { struct nfscllayout *lyp; struct nfsclflayout *flp, *tflp; struct nfscldevinfo *dip; struct nfsclflayouthead flh; int error = 0, islocked, layoutlen, recalled, retonclose; nfsv4stateid_t stateid; struct nfsclsession *tsep; *lypp = NULL; /* * If lyp is returned non-NULL, there will be a refcnt (shared lock) * on it, iff flp != NULL or a lock (exclusive lock) on it iff * flp == NULL. */ lyp = nfscl_getlayout(nmp->nm_clp, nfhp->nfh_fh, nfhp->nfh_len, off, &flp, &recalled); islocked = 0; if (lyp == NULL || flp == NULL) { if (recalled != 0) return (EIO); LIST_INIT(&flh); tsep = nfsmnt_mdssession(nmp); layoutlen = tsep->nfsess_maxcache - (NFSX_STATEID + 3 * NFSX_UNSIGNED); if (lyp == NULL) { stateid.seqid = 0; stateid.other[0] = stateidp->other[0]; stateid.other[1] = stateidp->other[1]; stateid.other[2] = stateidp->other[2]; error = nfsrpc_layoutget(nmp, nfhp->nfh_fh, nfhp->nfh_len, iomode, (uint64_t)0, UINT64_MAX, (uint64_t)0, layoutlen, &stateid, &retonclose, &flh, cred, p, NULL); } else { islocked = 1; stateid.seqid = lyp->nfsly_stateid.seqid; stateid.other[0] = lyp->nfsly_stateid.other[0]; stateid.other[1] = lyp->nfsly_stateid.other[1]; stateid.other[2] = lyp->nfsly_stateid.other[2]; error = nfsrpc_layoutget(nmp, nfhp->nfh_fh, nfhp->nfh_len, iomode, off, UINT64_MAX, (uint64_t)0, layoutlen, &stateid, &retonclose, &flh, cred, p, NULL); } if (error == 0) LIST_FOREACH(tflp, &flh, nfsfl_list) { error = nfscl_adddevinfo(nmp, NULL, tflp); if (error != 0) { error = nfsrpc_getdeviceinfo(nmp, tflp->nfsfl_dev, NFSLAYOUT_NFSV4_1_FILES, notifybitsp, &dip, cred, p); if (error != 0) break; error = nfscl_adddevinfo(nmp, dip, tflp); if (error != 0) printf( "getlayout: cannot add\n"); } } if (error == 0) { /* * nfscl_layout() always returns with the nfsly_lock * set to a refcnt (shared lock). */ error = nfscl_layout(nmp, vp, nfhp->nfh_fh, nfhp->nfh_len, &stateid, retonclose, &flh, &lyp, cred, p); if (error == 0) *lypp = lyp; } else if (islocked != 0) nfsv4_unlock(&lyp->nfsly_lock, 0); } else *lypp = lyp; return (error); } /* * Do a TCP connection plus exchange id and create session. * If successful, a "struct nfsclds" is linked into the list for the * mount point and a pointer to it is returned. */ static int nfsrpc_fillsa(struct nfsmount *nmp, struct sockaddr_storage *ssp, struct nfsclds **dspp, NFSPROC_T *p) { struct sockaddr_in *msad, *sad, *ssd; struct sockaddr_in6 *msad6, *sad6, *ssd6; struct nfsclclient *clp; struct nfssockreq *nrp; struct nfsclds *dsp, *tdsp; int error; enum nfsclds_state retv; uint32_t sequenceid; KASSERT(nmp->nm_sockreq.nr_cred != NULL, ("nfsrpc_fillsa: NULL nr_cred")); NFSLOCKCLSTATE(); clp = nmp->nm_clp; NFSUNLOCKCLSTATE(); if (clp == NULL) return (EPERM); if (ssp->ss_family == AF_INET) { ssd = (struct sockaddr_in *)ssp; NFSLOCKMNT(nmp); /* * Check to see if we already have a session for this * address that is usable for a DS. * Note that the MDS's address is in a different place * than the sessions already acquired for DS's. */ msad = (struct sockaddr_in *)nmp->nm_sockreq.nr_nam; tdsp = TAILQ_FIRST(&nmp->nm_sess); while (tdsp != NULL) { if (msad != NULL && msad->sin_family == AF_INET && ssd->sin_addr.s_addr == msad->sin_addr.s_addr && ssd->sin_port == msad->sin_port && (tdsp->nfsclds_flags & NFSCLDS_DS) != 0 && tdsp->nfsclds_sess.nfsess_defunct == 0) { *dspp = tdsp; NFSUNLOCKMNT(nmp); NFSCL_DEBUG(4, "fnd same addr\n"); return (0); } tdsp = TAILQ_NEXT(tdsp, nfsclds_list); if (tdsp != NULL && tdsp->nfsclds_sockp != NULL) msad = (struct sockaddr_in *) tdsp->nfsclds_sockp->nr_nam; else msad = NULL; } NFSUNLOCKMNT(nmp); /* No IP address match, so look for new/trunked one. */ sad = malloc(sizeof(*sad), M_SONAME, M_WAITOK | M_ZERO); sad->sin_len = sizeof(*sad); sad->sin_family = AF_INET; sad->sin_port = ssd->sin_port; sad->sin_addr.s_addr = ssd->sin_addr.s_addr; nrp = malloc(sizeof(*nrp), M_NFSSOCKREQ, M_WAITOK | M_ZERO); nrp->nr_nam = (struct sockaddr *)sad; } else if (ssp->ss_family == AF_INET6) { ssd6 = (struct sockaddr_in6 *)ssp; NFSLOCKMNT(nmp); /* * Check to see if we already have a session for this * address that is usable for a DS. * Note that the MDS's address is in a different place * than the sessions already acquired for DS's. */ msad6 = (struct sockaddr_in6 *)nmp->nm_sockreq.nr_nam; tdsp = TAILQ_FIRST(&nmp->nm_sess); while (tdsp != NULL) { if (msad6 != NULL && msad6->sin6_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(&ssd6->sin6_addr, &msad6->sin6_addr) && ssd6->sin6_port == msad6->sin6_port && (tdsp->nfsclds_flags & NFSCLDS_DS) != 0 && tdsp->nfsclds_sess.nfsess_defunct == 0) { *dspp = tdsp; NFSUNLOCKMNT(nmp); return (0); } tdsp = TAILQ_NEXT(tdsp, nfsclds_list); if (tdsp != NULL && tdsp->nfsclds_sockp != NULL) msad6 = (struct sockaddr_in6 *) tdsp->nfsclds_sockp->nr_nam; else msad6 = NULL; } NFSUNLOCKMNT(nmp); /* No IP address match, so look for new/trunked one. */ sad6 = malloc(sizeof(*sad6), M_SONAME, M_WAITOK | M_ZERO); sad6->sin6_len = sizeof(*sad6); sad6->sin6_family = AF_INET6; sad6->sin6_port = ssd6->sin6_port; NFSBCOPY(&ssd6->sin6_addr, &sad6->sin6_addr, sizeof(struct in6_addr)); nrp = malloc(sizeof(*nrp), M_NFSSOCKREQ, M_WAITOK | M_ZERO); nrp->nr_nam = (struct sockaddr *)sad6; } else return (EPERM); nrp->nr_sotype = SOCK_STREAM; mtx_init(&nrp->nr_mtx, "nfssock", NULL, MTX_DEF); nrp->nr_prog = NFS_PROG; nrp->nr_vers = NFS_VER4; /* * Use the credentials that were used for the mount, which are * in nmp->nm_sockreq.nr_cred for newnfs_connect() etc. * Ref. counting the credentials with crhold() is probably not * necessary, since nm_sockreq.nr_cred won't be crfree()'d until * unmount, but I did it anyhow. */ nrp->nr_cred = crhold(nmp->nm_sockreq.nr_cred); error = newnfs_connect(nmp, nrp, NULL, p, 0); NFSCL_DEBUG(3, "DS connect=%d\n", error); /* Now, do the exchangeid and create session. */ if (error == 0) { error = nfsrpc_exchangeid(nmp, clp, nrp, NFSV4EXCH_USEPNFSDS, &dsp, nrp->nr_cred, p); NFSCL_DEBUG(3, "DS exchangeid=%d\n", error); if (error != 0) newnfs_disconnect(nrp); } if (error == 0) { dsp->nfsclds_sockp = nrp; NFSLOCKMNT(nmp); retv = nfscl_getsameserver(nmp, dsp, &tdsp); NFSCL_DEBUG(3, "getsame ret=%d\n", retv); if (retv == NFSDSP_USETHISSESSION) { NFSUNLOCKMNT(nmp); /* * If there is already a session for this server, * use it. */ (void)newnfs_disconnect(nrp); nfscl_freenfsclds(dsp); *dspp = tdsp; return (0); } if (retv == NFSDSP_SEQTHISSESSION) sequenceid = tdsp->nfsclds_sess.nfsess_sequenceid; else sequenceid = dsp->nfsclds_sess.nfsess_sequenceid; NFSUNLOCKMNT(nmp); error = nfsrpc_createsession(nmp, &dsp->nfsclds_sess, nrp, sequenceid, 0, nrp->nr_cred, p); NFSCL_DEBUG(3, "DS createsess=%d\n", error); } else { NFSFREECRED(nrp->nr_cred); NFSFREEMUTEX(&nrp->nr_mtx); free(nrp->nr_nam, M_SONAME); free(nrp, M_NFSSOCKREQ); } if (error == 0) { NFSCL_DEBUG(3, "add DS session\n"); /* * Put it at the end of the list. That way the list * is ordered by when the entry was added. This matters * since the one done first is the one that should be * used for sequencid'ing any subsequent create sessions. */ NFSLOCKMNT(nmp); TAILQ_INSERT_TAIL(&nmp->nm_sess, dsp, nfsclds_list); NFSUNLOCKMNT(nmp); *dspp = dsp; } else if (dsp != NULL) { newnfs_disconnect(nrp); nfscl_freenfsclds(dsp); } return (error); } /* * Do the NFSv4.1 Reclaim Complete. */ int nfsrpc_reclaimcomplete(struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; int error; nfscl_reqstart(nd, NFSPROC_RECLAIMCOMPL, nmp, NULL, 0, NULL, NULL); NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED); *tl = newnfs_false; nd->nd_flag |= ND_USEGSSNAME; error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL); if (error != 0) return (error); error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } /* * Initialize the slot tables for a session. */ static void nfscl_initsessionslots(struct nfsclsession *sep) { int i; for (i = 0; i < NFSV4_CBSLOTS; i++) { if (sep->nfsess_cbslots[i].nfssl_reply != NULL) m_freem(sep->nfsess_cbslots[i].nfssl_reply); NFSBZERO(&sep->nfsess_cbslots[i], sizeof(struct nfsslot)); } for (i = 0; i < 64; i++) sep->nfsess_slotseq[i] = 0; sep->nfsess_slots = 0; } /* * Called to try and do an I/O operation via an NFSv4.1 Data Server (DS). */ int nfscl_doiods(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, uint32_t rwaccess, struct ucred *cred, NFSPROC_T *p) { struct nfsnode *np = VTONFS(vp); struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfscllayout *layp; struct nfscldevinfo *dip; struct nfsclflayout *rflp; nfsv4stateid_t stateid; struct ucred *newcred; uint64_t lastbyte, len, off, oresid, xfer; int eof, error, iolaymode, recalled; void *lckp; if (!NFSHASPNFS(nmp) || nfscl_enablecallb == 0 || nfs_numnfscbd == 0 || (np->n_flag & NNOLAYOUT) != 0) return (EIO); /* Now, get a reference cnt on the clientid for this mount. */ if (nfscl_getref(nmp) == 0) return (EIO); /* Find an appropriate stateid. */ newcred = NFSNEWCRED(cred); error = nfscl_getstateid(vp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, rwaccess, 1, newcred, p, &stateid, &lckp); if (error != 0) { NFSFREECRED(newcred); nfscl_relref(nmp); return (error); } /* Search for a layout for this file. */ off = uiop->uio_offset; layp = nfscl_getlayout(nmp->nm_clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len, off, &rflp, &recalled); if (layp == NULL || rflp == NULL) { if (recalled != 0) { NFSFREECRED(newcred); nfscl_relref(nmp); return (EIO); } if (layp != NULL) { nfscl_rellayout(layp, (rflp == NULL) ? 1 : 0); layp = NULL; } /* Try and get a Layout, if it is supported. */ if (rwaccess == NFSV4OPEN_ACCESSWRITE || (np->n_flag & NWRITEOPENED) != 0) iolaymode = NFSLAYOUTIOMODE_RW; else iolaymode = NFSLAYOUTIOMODE_READ; error = nfsrpc_getlayout(nmp, vp, np->n_fhp, iolaymode, NULL, &stateid, off, &layp, newcred, p); if (error != 0) { NFSLOCKNODE(np); np->n_flag |= NNOLAYOUT; NFSUNLOCKNODE(np); if (lckp != NULL) nfscl_lockderef(lckp); NFSFREECRED(newcred); if (layp != NULL) nfscl_rellayout(layp, 0); nfscl_relref(nmp); return (error); } } /* * Loop around finding a layout that works for the first part of * this I/O operation, and then call the function that actually * does the RPC. */ eof = 0; len = (uint64_t)uiop->uio_resid; while (len > 0 && error == 0 && eof == 0) { off = uiop->uio_offset; error = nfscl_findlayoutforio(layp, off, rwaccess, &rflp); if (error == 0) { oresid = xfer = (uint64_t)uiop->uio_resid; if (xfer > (rflp->nfsfl_end - rflp->nfsfl_off)) xfer = rflp->nfsfl_end - rflp->nfsfl_off; dip = nfscl_getdevinfo(nmp->nm_clp, rflp->nfsfl_dev, rflp->nfsfl_devp); if (dip != NULL) { error = nfscl_doflayoutio(vp, uiop, iomode, must_commit, &eof, &stateid, rwaccess, dip, layp, rflp, off, xfer, newcred, p); nfscl_reldevinfo(dip); lastbyte = off + xfer - 1; if (error == 0) { NFSLOCKCLSTATE(); if (lastbyte > layp->nfsly_lastbyte) layp->nfsly_lastbyte = lastbyte; NFSUNLOCKCLSTATE(); } else if (error == NFSERR_OPENMODE && rwaccess == NFSV4OPEN_ACCESSREAD) { NFSLOCKMNT(nmp); nmp->nm_state |= NFSSTA_OPENMODE; NFSUNLOCKMNT(nmp); } } else error = EIO; if (error == 0) len -= (oresid - (uint64_t)uiop->uio_resid); } } if (lckp != NULL) nfscl_lockderef(lckp); NFSFREECRED(newcred); nfscl_rellayout(layp, 0); nfscl_relref(nmp); return (error); } /* * Find a file layout that will handle the first bytes of the requested * range and return the information from it needed to to the I/O operation. */ int nfscl_findlayoutforio(struct nfscllayout *lyp, uint64_t off, uint32_t rwaccess, struct nfsclflayout **retflpp) { struct nfsclflayout *flp, *nflp, *rflp; uint32_t rw; rflp = NULL; rw = rwaccess; /* For reading, do the Read list first and then the Write list. */ do { if (rw == NFSV4OPEN_ACCESSREAD) flp = LIST_FIRST(&lyp->nfsly_flayread); else flp = LIST_FIRST(&lyp->nfsly_flayrw); while (flp != NULL) { nflp = LIST_NEXT(flp, nfsfl_list); if (flp->nfsfl_off > off) break; if (flp->nfsfl_end > off && (rflp == NULL || rflp->nfsfl_end < flp->nfsfl_end)) rflp = flp; flp = nflp; } if (rw == NFSV4OPEN_ACCESSREAD) rw = NFSV4OPEN_ACCESSWRITE; else rw = 0; } while (rw != 0); if (rflp != NULL) { /* This one covers the most bytes starting at off. */ *retflpp = rflp; return (0); } return (EIO); } /* * Do I/O using an NFSv4.1 file layout. */ static int nfscl_doflayoutio(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, int *eofp, nfsv4stateid_t *stateidp, int rwflag, struct nfscldevinfo *dp, struct nfscllayout *lyp, struct nfsclflayout *flp, uint64_t off, uint64_t len, struct ucred *cred, NFSPROC_T *p) { uint64_t io_off, rel_off, stripe_unit_size, transfer, xfer; int commit_thru_mds, error = 0, stripe_index, stripe_pos; struct nfsnode *np; struct nfsfh *fhp; struct nfsclds **dspp; np = VTONFS(vp); rel_off = off - flp->nfsfl_patoff; stripe_unit_size = (flp->nfsfl_util >> 6) & 0x3ffffff; stripe_pos = (rel_off / stripe_unit_size + flp->nfsfl_stripe1) % dp->nfsdi_stripecnt; transfer = stripe_unit_size - (rel_off % stripe_unit_size); /* Loop around, doing I/O for each stripe unit. */ while (len > 0 && error == 0) { stripe_index = nfsfldi_stripeindex(dp, stripe_pos); dspp = nfsfldi_addr(dp, stripe_index); if (len > transfer) xfer = transfer; else xfer = len; if ((flp->nfsfl_util & NFSFLAYUTIL_DENSE) != 0) { /* Dense layout. */ if (stripe_pos >= flp->nfsfl_fhcnt) return (EIO); fhp = flp->nfsfl_fh[stripe_pos]; io_off = (rel_off / (stripe_unit_size * dp->nfsdi_stripecnt)) * stripe_unit_size + rel_off % stripe_unit_size; } else { /* Sparse layout. */ if (flp->nfsfl_fhcnt > 1) { if (stripe_index >= flp->nfsfl_fhcnt) return (EIO); fhp = flp->nfsfl_fh[stripe_index]; } else if (flp->nfsfl_fhcnt == 1) fhp = flp->nfsfl_fh[0]; else fhp = np->n_fhp; io_off = off; } if ((flp->nfsfl_util & NFSFLAYUTIL_COMMIT_THRU_MDS) != 0) commit_thru_mds = 1; else commit_thru_mds = 0; if (rwflag == FREAD) error = nfsrpc_readds(vp, uiop, stateidp, eofp, *dspp, io_off, xfer, fhp, cred, p); else { error = nfsrpc_writeds(vp, uiop, iomode, must_commit, stateidp, *dspp, io_off, xfer, fhp, commit_thru_mds, cred, p); if (error == 0) { NFSLOCKCLSTATE(); lyp->nfsly_flags |= NFSLY_WRITTEN; NFSUNLOCKCLSTATE(); } } if (error == 0) { transfer = stripe_unit_size; stripe_pos = (stripe_pos + 1) % dp->nfsdi_stripecnt; len -= xfer; off += xfer; } } return (error); } /* * The actual read RPC done to a DS. */ static int nfsrpc_readds(vnode_t vp, struct uio *uiop, nfsv4stateid_t *stateidp, int *eofp, struct nfsclds *dsp, uint64_t io_off, int len, struct nfsfh *fhp, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; int error, retlen; struct nfsrv_descript nfsd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfsrv_descript *nd = &nfsd; struct nfssockreq *nrp; nd->nd_mrep = NULL; nfscl_reqstart(nd, NFSPROC_READDS, nmp, fhp->nfh_fh, fhp->nfh_len, NULL, &dsp->nfsclds_sess); nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSEQIDZERO); NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3); txdr_hyper(io_off, tl); *(tl + 2) = txdr_unsigned(len); nrp = dsp->nfsclds_sockp; if (nrp == NULL) /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); if (error != 0) return (error); if (nd->nd_repstat != 0) { error = nd->nd_repstat; goto nfsmout; } NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED); *eofp = fxdr_unsigned(int, *tl); NFSM_STRSIZ(retlen, len); error = nfsm_mbufuio(nd, uiop, retlen); nfsmout: if (nd->nd_mrep != NULL) mbuf_freem(nd->nd_mrep); return (error); } /* * The actual write RPC done to a DS. */ static int nfsrpc_writeds(vnode_t vp, struct uio *uiop, int *iomode, int *must_commit, nfsv4stateid_t *stateidp, struct nfsclds *dsp, uint64_t io_off, int len, struct nfsfh *fhp, int commit_thru_mds, struct ucred *cred, NFSPROC_T *p) { uint32_t *tl; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); int error, rlen, commit, committed = NFSWRITE_FILESYNC; int32_t backup; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; struct nfssockreq *nrp; KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1")); nd->nd_mrep = NULL; nfscl_reqstart(nd, NFSPROC_WRITEDS, nmp, fhp->nfh_fh, fhp->nfh_len, NULL, &dsp->nfsclds_sess); nfsm_stateidtom(nd, stateidp, NFSSTATEID_PUTSEQIDZERO); NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED); txdr_hyper(io_off, tl); tl += 2; *tl++ = txdr_unsigned(*iomode); *tl = txdr_unsigned(len); nfsm_uiombuf(nd, uiop, len); nrp = dsp->nfsclds_sockp; if (nrp == NULL) /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); if (error != 0) return (error); if (nd->nd_repstat != 0) { /* * In case the rpc gets retried, roll * the uio fileds changed by nfsm_uiombuf() * back. */ uiop->uio_offset -= len; uio_uio_resid_add(uiop, len); uio_iov_base_add(uiop, -len); uio_iov_len_add(uiop, len); error = nd->nd_repstat; } else { NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF); rlen = fxdr_unsigned(int, *tl++); if (rlen == 0) { error = NFSERR_IO; goto nfsmout; } else if (rlen < len) { backup = len - rlen; uio_iov_base_add(uiop, -(backup)); uio_iov_len_add(uiop, backup); uiop->uio_offset -= backup; uio_uio_resid_add(uiop, backup); len = rlen; } commit = fxdr_unsigned(int, *tl++); /* * Return the lowest commitment level * obtained by any of the RPCs. */ if (committed == NFSWRITE_FILESYNC) committed = commit; else if (committed == NFSWRITE_DATASYNC && commit == NFSWRITE_UNSTABLE) committed = commit; if (commit_thru_mds != 0) { NFSLOCKMNT(nmp); if (!NFSHASWRITEVERF(nmp)) { NFSBCOPY(tl, nmp->nm_verf, NFSX_VERF); NFSSETWRITEVERF(nmp); } else if (NFSBCMP(tl, nmp->nm_verf, NFSX_VERF)) { *must_commit = 1; NFSBCOPY(tl, nmp->nm_verf, NFSX_VERF); } NFSUNLOCKMNT(nmp); } else { NFSLOCKDS(dsp); if ((dsp->nfsclds_flags & NFSCLDS_HASWRITEVERF) == 0) { NFSBCOPY(tl, dsp->nfsclds_verf, NFSX_VERF); dsp->nfsclds_flags |= NFSCLDS_HASWRITEVERF; } else if (NFSBCMP(tl, dsp->nfsclds_verf, NFSX_VERF)) { *must_commit = 1; NFSBCOPY(tl, dsp->nfsclds_verf, NFSX_VERF); } NFSUNLOCKDS(dsp); } } nfsmout: if (nd->nd_mrep != NULL) mbuf_freem(nd->nd_mrep); *iomode = committed; if (nd->nd_repstat != 0 && error == 0) error = nd->nd_repstat; return (error); } /* * Free up the nfsclds structure. */ void nfscl_freenfsclds(struct nfsclds *dsp) { int i; if (dsp == NULL) return; if (dsp->nfsclds_sockp != NULL) { NFSFREECRED(dsp->nfsclds_sockp->nr_cred); NFSFREEMUTEX(&dsp->nfsclds_sockp->nr_mtx); free(dsp->nfsclds_sockp->nr_nam, M_SONAME); free(dsp->nfsclds_sockp, M_NFSSOCKREQ); } NFSFREEMUTEX(&dsp->nfsclds_mtx); NFSFREEMUTEX(&dsp->nfsclds_sess.nfsess_mtx); for (i = 0; i < NFSV4_CBSLOTS; i++) { if (dsp->nfsclds_sess.nfsess_cbslots[i].nfssl_reply != NULL) m_freem( dsp->nfsclds_sess.nfsess_cbslots[i].nfssl_reply); } free(dsp, M_NFSCLDS); } static enum nfsclds_state nfscl_getsameserver(struct nfsmount *nmp, struct nfsclds *newdsp, struct nfsclds **retdspp) { struct nfsclds *dsp, *cur_dsp; /* * Search the list of nfsclds structures for one with the same * server. */ cur_dsp = NULL; TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) { if (dsp->nfsclds_servownlen == newdsp->nfsclds_servownlen && dsp->nfsclds_servownlen != 0 && !NFSBCMP(dsp->nfsclds_serverown, newdsp->nfsclds_serverown, dsp->nfsclds_servownlen) && dsp->nfsclds_sess.nfsess_defunct == 0) { NFSCL_DEBUG(4, "fnd same fdsp=%p dsp=%p flg=0x%x\n", TAILQ_FIRST(&nmp->nm_sess), dsp, dsp->nfsclds_flags); /* Server major id matches. */ if ((dsp->nfsclds_flags & NFSCLDS_DS) != 0) { *retdspp = dsp; return (NFSDSP_USETHISSESSION); } /* * Note the first match, so it can be used for * sequence'ing new sessions. */ if (cur_dsp == NULL) cur_dsp = dsp; } } if (cur_dsp != NULL) { *retdspp = cur_dsp; return (NFSDSP_SEQTHISSESSION); } return (NFSDSP_NOTFOUND); } #ifdef notyet /* * NFS commit rpc to a DS. */ static int nfsrpc_commitds(vnode_t vp, uint64_t offset, int cnt, struct nfsclds *dsp, struct nfsfh *fhp, struct ucred *cred, NFSPROC_T *p, void *stuff) { uint32_t *tl; struct nfsrv_descript nfsd, *nd = &nfsd; struct nfsmount *nmp = VFSTONFS(vnode_mount(vp)); struct nfssockreq *nrp; int error; nfscl_reqstart(nd, NFSPROC_COMMITDS, nmp, fhp->nfh_fh, fhp->nfh_len, NULL, &dsp->nfsclds_sess); NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED); txdr_hyper(offset, tl); tl += 2; *tl = txdr_unsigned(cnt); nrp = dsp->nfsclds_sockp; if (nrp == NULL) /* If NULL, use the MDS socket. */ nrp = &nmp->nm_sockreq; error = newnfs_request(nd, nmp, NULL, nrp, vp, p, cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, &dsp->nfsclds_sess); if (error) return (error); if (nd->nd_repstat == 0) { NFSM_DISSECT(tl, u_int32_t *, NFSX_VERF); NFSLOCKDS(dsp); if (NFSBCMP(tl, dsp->nfsclds_verf, NFSX_VERF)) { NFSBCOPY(tl, dsp->nfsclds_verf, NFSX_VERF); error = NFSERR_STALEWRITEVERF; } NFSUNLOCKDS(dsp); } nfsmout: if (error == 0 && nd->nd_repstat != 0) error = nd->nd_repstat; mbuf_freem(nd->nd_mrep); return (error); } #endif Index: head/sys/kern/vfs_bio.c =================================================================== --- head/sys/kern/vfs_bio.c (revision 320061) +++ head/sys/kern/vfs_bio.c (revision 320062) @@ -1,5003 +1,5033 @@ /*- * Copyright (c) 2004 Poul-Henning Kamp * Copyright (c) 1994,1997 John S. Dyson * Copyright (c) 2013 The FreeBSD Foundation * All rights reserved. * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * 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. */ /* * this file contains a new buffer I/O scheme implementing a coherent * VM object and buffer cache scheme. Pains have been taken to make * sure that the performance degradation associated with schemes such * as this is not realized. * * Author: John S. Dyson * Significant help during the development and debugging phases * had been provided by David Greenman, also of the FreeBSD core team. * * see man buf(9) for more info. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_compat.h" #include "opt_swap.h" static MALLOC_DEFINE(M_BIOBUF, "biobuf", "BIO buffer"); struct bio_ops bioops; /* I/O operation notification */ struct buf_ops buf_ops_bio = { .bop_name = "buf_ops_bio", .bop_write = bufwrite, .bop_strategy = bufstrategy, .bop_sync = bufsync, .bop_bdflush = bufbdflush, }; static struct buf *buf; /* buffer header pool */ extern struct buf *swbuf; /* Swap buffer header pool. */ caddr_t unmapped_buf; /* Used below and for softdep flushing threads in ufs/ffs/ffs_softdep.c */ struct proc *bufdaemonproc; struct proc *bufspacedaemonproc; static int inmem(struct vnode *vp, daddr_t blkno); static void vm_hold_free_pages(struct buf *bp, int newbsize); static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_clean_pages_dirty_buf(struct buf *bp); static void vfs_setdirty_locked_object(struct buf *bp); static void vfs_vmio_invalidate(struct buf *bp); static void vfs_vmio_truncate(struct buf *bp, int npages); static void vfs_vmio_extend(struct buf *bp, int npages, int size); static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno); static int buf_flush(struct vnode *vp, int); static int buf_recycle(bool); static int buf_scan(bool); static int flushbufqueues(struct vnode *, int, int); static void buf_daemon(void); static void bremfreel(struct buf *bp); static __inline void bd_wakeup(void); static int sysctl_runningspace(SYSCTL_HANDLER_ARGS); static void bufkva_reclaim(vmem_t *, int); static void bufkva_free(struct buf *); static int buf_import(void *, void **, int, int); static void buf_release(void *, void **, int); +static void maxbcachebuf_adjust(void); #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS); #endif int vmiodirenable = TRUE; SYSCTL_INT(_vfs, OID_AUTO, vmiodirenable, CTLFLAG_RW, &vmiodirenable, 0, "Use the VM system for directory writes"); long runningbufspace; SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0, "Amount of presently outstanding async buffer io"); static long bufspace; #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCTL_PROC(_vfs, OID_AUTO, bufspace, CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RD, &bufspace, 0, sysctl_bufspace, "L", "Virtual memory used for buffers"); #else SYSCTL_LONG(_vfs, OID_AUTO, bufspace, CTLFLAG_RD, &bufspace, 0, "Physical memory used for buffers"); #endif static long bufkvaspace; SYSCTL_LONG(_vfs, OID_AUTO, bufkvaspace, CTLFLAG_RD, &bufkvaspace, 0, "Kernel virtual memory used for buffers"); static long maxbufspace; SYSCTL_LONG(_vfs, OID_AUTO, maxbufspace, CTLFLAG_RW, &maxbufspace, 0, "Maximum allowed value of bufspace (including metadata)"); static long bufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0, "Amount of malloced memory for buffers"); static long maxbufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RW, &maxbufmallocspace, 0, "Maximum amount of malloced memory for buffers"); static long lobufspace; SYSCTL_LONG(_vfs, OID_AUTO, lobufspace, CTLFLAG_RW, &lobufspace, 0, "Minimum amount of buffers we want to have"); long hibufspace; SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RW, &hibufspace, 0, "Maximum allowed value of bufspace (excluding metadata)"); long bufspacethresh; SYSCTL_LONG(_vfs, OID_AUTO, bufspacethresh, CTLFLAG_RW, &bufspacethresh, 0, "Bufspace consumed before waking the daemon to free some"); static int buffreekvacnt; SYSCTL_INT(_vfs, OID_AUTO, buffreekvacnt, CTLFLAG_RW, &buffreekvacnt, 0, "Number of times we have freed the KVA space from some buffer"); static int bufdefragcnt; SYSCTL_INT(_vfs, OID_AUTO, bufdefragcnt, CTLFLAG_RW, &bufdefragcnt, 0, "Number of times we have had to repeat buffer allocation to defragment"); static long lorunningspace; SYSCTL_PROC(_vfs, OID_AUTO, lorunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &lorunningspace, 0, sysctl_runningspace, "L", "Minimum preferred space used for in-progress I/O"); static long hirunningspace; SYSCTL_PROC(_vfs, OID_AUTO, hirunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &hirunningspace, 0, sysctl_runningspace, "L", "Maximum amount of space to use for in-progress I/O"); int dirtybufferflushes; SYSCTL_INT(_vfs, OID_AUTO, dirtybufferflushes, CTLFLAG_RW, &dirtybufferflushes, 0, "Number of bdwrite to bawrite conversions to limit dirty buffers"); int bdwriteskip; SYSCTL_INT(_vfs, OID_AUTO, bdwriteskip, CTLFLAG_RW, &bdwriteskip, 0, "Number of buffers supplied to bdwrite with snapshot deadlock risk"); int altbufferflushes; SYSCTL_INT(_vfs, OID_AUTO, altbufferflushes, CTLFLAG_RW, &altbufferflushes, 0, "Number of fsync flushes to limit dirty buffers"); static int recursiveflushes; SYSCTL_INT(_vfs, OID_AUTO, recursiveflushes, CTLFLAG_RW, &recursiveflushes, 0, "Number of flushes skipped due to being recursive"); static int numdirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, numdirtybuffers, CTLFLAG_RD, &numdirtybuffers, 0, "Number of buffers that are dirty (has unwritten changes) at the moment"); static int lodirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, lodirtybuffers, CTLFLAG_RW, &lodirtybuffers, 0, "How many buffers we want to have free before bufdaemon can sleep"); static int hidirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, hidirtybuffers, CTLFLAG_RW, &hidirtybuffers, 0, "When the number of dirty buffers is considered severe"); int dirtybufthresh; SYSCTL_INT(_vfs, OID_AUTO, dirtybufthresh, CTLFLAG_RW, &dirtybufthresh, 0, "Number of bdwrite to bawrite conversions to clear dirty buffers"); static int numfreebuffers; SYSCTL_INT(_vfs, OID_AUTO, numfreebuffers, CTLFLAG_RD, &numfreebuffers, 0, "Number of free buffers"); static int lofreebuffers; SYSCTL_INT(_vfs, OID_AUTO, lofreebuffers, CTLFLAG_RW, &lofreebuffers, 0, "Target number of free buffers"); static int hifreebuffers; SYSCTL_INT(_vfs, OID_AUTO, hifreebuffers, CTLFLAG_RW, &hifreebuffers, 0, "Threshold for clean buffer recycling"); static int getnewbufcalls; SYSCTL_INT(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RW, &getnewbufcalls, 0, "Number of calls to getnewbuf"); static int getnewbufrestarts; SYSCTL_INT(_vfs, OID_AUTO, getnewbufrestarts, CTLFLAG_RW, &getnewbufrestarts, 0, "Number of times getnewbuf has had to restart a buffer acquisition"); static int mappingrestarts; SYSCTL_INT(_vfs, OID_AUTO, mappingrestarts, CTLFLAG_RW, &mappingrestarts, 0, "Number of times getblk has had to restart a buffer mapping for " "unmapped buffer"); static int numbufallocfails; SYSCTL_INT(_vfs, OID_AUTO, numbufallocfails, CTLFLAG_RW, &numbufallocfails, 0, "Number of times buffer allocations failed"); static int flushbufqtarget = 100; SYSCTL_INT(_vfs, OID_AUTO, flushbufqtarget, CTLFLAG_RW, &flushbufqtarget, 0, "Amount of work to do in flushbufqueues when helping bufdaemon"); static long notbufdflushes; SYSCTL_LONG(_vfs, OID_AUTO, notbufdflushes, CTLFLAG_RD, ¬bufdflushes, 0, "Number of dirty buffer flushes done by the bufdaemon helpers"); static long barrierwrites; SYSCTL_LONG(_vfs, OID_AUTO, barrierwrites, CTLFLAG_RW, &barrierwrites, 0, "Number of barrier writes"); SYSCTL_INT(_vfs, OID_AUTO, unmapped_buf_allowed, CTLFLAG_RD, &unmapped_buf_allowed, 0, "Permit the use of the unmapped i/o"); +int maxbcachebuf = MAXBCACHEBUF; +SYSCTL_INT(_vfs, OID_AUTO, maxbcachebuf, CTLFLAG_RDTUN, &maxbcachebuf, 0, + "Maximum size of a buffer cache block"); /* * This lock synchronizes access to bd_request. */ static struct mtx_padalign bdlock; /* * This lock protects the runningbufreq and synchronizes runningbufwakeup and * waitrunningbufspace(). */ static struct mtx_padalign rbreqlock; /* * Lock that protects needsbuffer and the sleeps/wakeups surrounding it. */ static struct rwlock_padalign nblock; /* * Lock that protects bdirtywait. */ static struct mtx_padalign bdirtylock; /* * Wakeup point for bufdaemon, as well as indicator of whether it is already * active. Set to 1 when the bufdaemon is already "on" the queue, 0 when it * is idling. */ static int bd_request; /* * Request/wakeup point for the bufspace daemon. */ static int bufspace_request; /* * Request for the buf daemon to write more buffers than is indicated by * lodirtybuf. This may be necessary to push out excess dependencies or * defragment the address space where a simple count of the number of dirty * buffers is insufficient to characterize the demand for flushing them. */ static int bd_speedupreq; /* * Synchronization (sleep/wakeup) variable for active buffer space requests. * Set when wait starts, cleared prior to wakeup(). * Used in runningbufwakeup() and waitrunningbufspace(). */ static int runningbufreq; /* * Synchronization (sleep/wakeup) variable for buffer requests. * Can contain the VFS_BIO_NEED flags defined below; setting/clearing is done * by and/or. * Used in numdirtywakeup(), bufspace_wakeup(), bwillwrite(), * getnewbuf(), and getblk(). */ static volatile int needsbuffer; /* * Synchronization for bwillwrite() waiters. */ static int bdirtywait; /* * Definitions for the buffer free lists. */ #define QUEUE_NONE 0 /* on no queue */ #define QUEUE_EMPTY 1 /* empty buffer headers */ #define QUEUE_DIRTY 2 /* B_DELWRI buffers */ #define QUEUE_CLEAN 3 /* non-B_DELWRI buffers */ #define QUEUE_SENTINEL 1024 /* not an queue index, but mark for sentinel */ /* Maximum number of clean buffer queues. */ #define CLEAN_QUEUES 16 /* Configured number of clean queues. */ static int clean_queues; /* Maximum number of buffer queues. */ #define BUFFER_QUEUES (QUEUE_CLEAN + CLEAN_QUEUES) /* Queues for free buffers with various properties */ static TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES] = { { 0 } }; #ifdef INVARIANTS static int bq_len[BUFFER_QUEUES]; #endif /* * Lock for each bufqueue */ static struct mtx_padalign bqlocks[BUFFER_QUEUES]; /* * per-cpu empty buffer cache. */ uma_zone_t buf_zone; /* * Single global constant for BUF_WMESG, to avoid getting multiple references. * buf_wmesg is referred from macros. */ const char *buf_wmesg = BUF_WMESG; static int sysctl_runningspace(SYSCTL_HANDLER_ARGS) { long value; int error; value = *(long *)arg1; error = sysctl_handle_long(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); mtx_lock(&rbreqlock); if (arg1 == &hirunningspace) { if (value < lorunningspace) error = EINVAL; else hirunningspace = value; } else { KASSERT(arg1 == &lorunningspace, ("%s: unknown arg1", __func__)); if (value > hirunningspace) error = EINVAL; else lorunningspace = value; } mtx_unlock(&rbreqlock); return (error); } #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS) { long lvalue; int ivalue; if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long)) return (sysctl_handle_long(oidp, arg1, arg2, req)); lvalue = *(long *)arg1; if (lvalue > INT_MAX) /* On overflow, still write out a long to trigger ENOMEM. */ return (sysctl_handle_long(oidp, &lvalue, 0, req)); ivalue = lvalue; return (sysctl_handle_int(oidp, &ivalue, 0, req)); } #endif static int bqcleanq(void) { static int nextq; return ((atomic_fetchadd_int(&nextq, 1) % clean_queues) + QUEUE_CLEAN); } static int bqisclean(int qindex) { return (qindex >= QUEUE_CLEAN && qindex < QUEUE_CLEAN + CLEAN_QUEUES); } /* * bqlock: * * Return the appropriate queue lock based on the index. */ static inline struct mtx * bqlock(int qindex) { return (struct mtx *)&bqlocks[qindex]; } /* * bdirtywakeup: * * Wakeup any bwillwrite() waiters. */ static void bdirtywakeup(void) { mtx_lock(&bdirtylock); if (bdirtywait) { bdirtywait = 0; wakeup(&bdirtywait); } mtx_unlock(&bdirtylock); } /* * bdirtysub: * * Decrement the numdirtybuffers count by one and wakeup any * threads blocked in bwillwrite(). */ static void bdirtysub(void) { if (atomic_fetchadd_int(&numdirtybuffers, -1) == (lodirtybuffers + hidirtybuffers) / 2) bdirtywakeup(); } /* * bdirtyadd: * * Increment the numdirtybuffers count by one and wakeup the buf * daemon if needed. */ static void bdirtyadd(void) { /* * Only do the wakeup once as we cross the boundary. The * buf daemon will keep running until the condition clears. */ if (atomic_fetchadd_int(&numdirtybuffers, 1) == (lodirtybuffers + hidirtybuffers) / 2) bd_wakeup(); } /* * bufspace_wakeup: * * Called when buffer space is potentially available for recovery. * getnewbuf() will block on this flag when it is unable to free * sufficient buffer space. Buffer space becomes recoverable when * bp's get placed back in the queues. */ static void bufspace_wakeup(void) { /* * If someone is waiting for bufspace, wake them up. * * Since needsbuffer is set prior to doing an additional queue * scan it is safe to check for the flag prior to acquiring the * lock. The thread that is preparing to scan again before * blocking would discover the buf we released. */ if (needsbuffer) { rw_rlock(&nblock); if (atomic_cmpset_int(&needsbuffer, 1, 0) == 1) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } } /* * bufspace_daemonwakeup: * * Wakeup the daemon responsible for freeing clean bufs. */ static void bufspace_daemonwakeup(void) { rw_rlock(&nblock); if (bufspace_request == 0) { bufspace_request = 1; wakeup(&bufspace_request); } rw_runlock(&nblock); } /* * bufspace_adjust: * * Adjust the reported bufspace for a KVA managed buffer, possibly * waking any waiters. */ static void bufspace_adjust(struct buf *bp, int bufsize) { long space; int diff; KASSERT((bp->b_flags & B_MALLOC) == 0, ("bufspace_adjust: malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) { atomic_subtract_long(&bufspace, -diff); bufspace_wakeup(); } else { space = atomic_fetchadd_long(&bufspace, diff); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + diff >= bufspacethresh) bufspace_daemonwakeup(); } bp->b_bufsize = bufsize; } /* * bufspace_reserve: * * Reserve bufspace before calling allocbuf(). metadata has a * different space limit than data. */ static int bufspace_reserve(int size, bool metadata) { long limit; long space; if (metadata) limit = maxbufspace; else limit = hibufspace; do { space = bufspace; if (space + size > limit) return (ENOSPC); } while (atomic_cmpset_long(&bufspace, space, space + size) == 0); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + size >= bufspacethresh) bufspace_daemonwakeup(); return (0); } /* * bufspace_release: * * Release reserved bufspace after bufspace_adjust() has consumed it. */ static void bufspace_release(int size) { atomic_subtract_long(&bufspace, size); bufspace_wakeup(); } /* * bufspace_wait: * * Wait for bufspace, acting as the buf daemon if a locked vnode is * supplied. needsbuffer must be set in a safe fashion prior to * polling for space. The operation must be re-tried on return. */ static void bufspace_wait(struct vnode *vp, int gbflags, int slpflag, int slptimeo) { struct thread *td; int error, fl, norunbuf; if ((gbflags & GB_NOWAIT_BD) != 0) return; td = curthread; rw_wlock(&nblock); while (needsbuffer != 0) { if (vp != NULL && vp->v_type != VCHR && (td->td_pflags & TDP_BUFNEED) == 0) { rw_wunlock(&nblock); /* * getblk() is called with a vnode locked, and * some majority of the dirty buffers may as * well belong to the vnode. Flushing the * buffers there would make a progress that * cannot be achieved by the buf_daemon, that * cannot lock the vnode. */ norunbuf = ~(TDP_BUFNEED | TDP_NORUNNINGBUF) | (td->td_pflags & TDP_NORUNNINGBUF); /* * Play bufdaemon. The getnewbuf() function * may be called while the thread owns lock * for another dirty buffer for the same * vnode, which makes it impossible to use * VOP_FSYNC() there, due to the buffer lock * recursion. */ td->td_pflags |= TDP_BUFNEED | TDP_NORUNNINGBUF; fl = buf_flush(vp, flushbufqtarget); td->td_pflags &= norunbuf; rw_wlock(&nblock); if (fl != 0) continue; if (needsbuffer == 0) break; } error = rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, (PRIBIO + 4) | slpflag, "newbuf", slptimeo); if (error != 0) break; } rw_wunlock(&nblock); } /* * bufspace_daemon: * * buffer space management daemon. Tries to maintain some marginal * amount of free buffer space so that requesting processes neither * block nor work to reclaim buffers. */ static void bufspace_daemon(void) { for (;;) { kproc_suspend_check(bufspacedaemonproc); /* * Free buffers from the clean queue until we meet our * targets. * * Theory of operation: The buffer cache is most efficient * when some free buffer headers and space are always * available to getnewbuf(). This daemon attempts to prevent * the excessive blocking and synchronization associated * with shortfall. It goes through three phases according * demand: * * 1) The daemon wakes up voluntarily once per-second * during idle periods when the counters are below * the wakeup thresholds (bufspacethresh, lofreebuffers). * * 2) The daemon wakes up as we cross the thresholds * ahead of any potential blocking. This may bounce * slightly according to the rate of consumption and * release. * * 3) The daemon and consumers are starved for working * clean buffers. This is the 'bufspace' sleep below * which will inefficiently trade bufs with bqrelse * until we return to condition 2. */ while (bufspace > lobufspace || numfreebuffers < hifreebuffers) { if (buf_recycle(false) != 0) { atomic_set_int(&needsbuffer, 1); if (buf_recycle(false) != 0) { rw_wlock(&nblock); if (needsbuffer) rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, PRIBIO|PDROP, "bufspace", hz/10); else rw_wunlock(&nblock); } } maybe_yield(); } /* * Re-check our limits under the exclusive nblock. */ rw_wlock(&nblock); if (bufspace < bufspacethresh && numfreebuffers > lofreebuffers) { bufspace_request = 0; rw_sleep(&bufspace_request, &nblock, PRIBIO|PDROP, "-", hz); } else rw_wunlock(&nblock); } } static struct kproc_desc bufspace_kp = { "bufspacedaemon", bufspace_daemon, &bufspacedaemonproc }; SYSINIT(bufspacedaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &bufspace_kp); /* * bufmallocadjust: * * Adjust the reported bufspace for a malloc managed buffer, possibly * waking any waiters. */ static void bufmallocadjust(struct buf *bp, int bufsize) { int diff; KASSERT((bp->b_flags & B_MALLOC) != 0, ("bufmallocadjust: non-malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) atomic_subtract_long(&bufmallocspace, -diff); else atomic_add_long(&bufmallocspace, diff); bp->b_bufsize = bufsize; } /* * runningwakeup: * * Wake up processes that are waiting on asynchronous writes to fall * below lorunningspace. */ static void runningwakeup(void) { mtx_lock(&rbreqlock); if (runningbufreq) { runningbufreq = 0; wakeup(&runningbufreq); } mtx_unlock(&rbreqlock); } /* * runningbufwakeup: * * Decrement the outstanding write count according. */ void runningbufwakeup(struct buf *bp) { long space, bspace; bspace = bp->b_runningbufspace; if (bspace == 0) return; space = atomic_fetchadd_long(&runningbufspace, -bspace); KASSERT(space >= bspace, ("runningbufspace underflow %ld %ld", space, bspace)); bp->b_runningbufspace = 0; /* * Only acquire the lock and wakeup on the transition from exceeding * the threshold to falling below it. */ if (space < lorunningspace) return; if (space - bspace > lorunningspace) return; runningwakeup(); } /* * waitrunningbufspace() * * runningbufspace is a measure of the amount of I/O currently * running. This routine is used in async-write situations to * prevent creating huge backups of pending writes to a device. * Only asynchronous writes are governed by this function. * * This does NOT turn an async write into a sync write. It waits * for earlier writes to complete and generally returns before the * caller's write has reached the device. */ void waitrunningbufspace(void) { mtx_lock(&rbreqlock); while (runningbufspace > hirunningspace) { runningbufreq = 1; msleep(&runningbufreq, &rbreqlock, PVM, "wdrain", 0); } mtx_unlock(&rbreqlock); } /* * vfs_buf_test_cache: * * Called when a buffer is extended. This function clears the B_CACHE * bit if the newly extended portion of the buffer does not contain * valid data. */ static __inline void vfs_buf_test_cache(struct buf *bp, vm_ooffset_t foff, vm_offset_t off, vm_offset_t size, vm_page_t m) { VM_OBJECT_ASSERT_LOCKED(m->object); if (bp->b_flags & B_CACHE) { int base = (foff + off) & PAGE_MASK; if (vm_page_is_valid(m, base, size) == 0) bp->b_flags &= ~B_CACHE; } } /* Wake up the buffer daemon if necessary */ static __inline void bd_wakeup(void) { mtx_lock(&bdlock); if (bd_request == 0) { bd_request = 1; wakeup(&bd_request); } mtx_unlock(&bdlock); } /* + * Adjust the maxbcachbuf tunable. + */ +static void +maxbcachebuf_adjust(void) +{ + int i; + + /* + * maxbcachebuf must be a power of 2 >= MAXBSIZE. + */ + i = 2; + while (i * 2 <= maxbcachebuf) + i *= 2; + maxbcachebuf = i; + if (maxbcachebuf < MAXBSIZE) + maxbcachebuf = MAXBSIZE; + if (maxbcachebuf > MAXPHYS) + maxbcachebuf = MAXPHYS; + if (bootverbose != 0 && maxbcachebuf != MAXBCACHEBUF) + printf("maxbcachebuf=%d\n", maxbcachebuf); +} + +/* * bd_speedup - speedup the buffer cache flushing code */ void bd_speedup(void) { int needwake; mtx_lock(&bdlock); needwake = 0; if (bd_speedupreq == 0 || bd_request == 0) needwake = 1; bd_speedupreq = 1; bd_request = 1; if (needwake) wakeup(&bd_request); mtx_unlock(&bdlock); } #ifndef NSWBUF_MIN #define NSWBUF_MIN 16 #endif #ifdef __i386__ #define TRANSIENT_DENOM 5 #else #define TRANSIENT_DENOM 10 #endif /* * Calculating buffer cache scaling values and reserve space for buffer * headers. This is called during low level kernel initialization and * may be called more then once. We CANNOT write to the memory area * being reserved at this time. */ caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est) { int tuned_nbuf; long maxbuf, maxbuf_sz, buf_sz, biotmap_sz; /* * physmem_est is in pages. Convert it to kilobytes (assumes * PAGE_SIZE is >= 1K) */ physmem_est = physmem_est * (PAGE_SIZE / 1024); + maxbcachebuf_adjust(); /* * The nominal buffer size (and minimum KVA allocation) is BKVASIZE. * For the first 64MB of ram nominally allocate sufficient buffers to * cover 1/4 of our ram. Beyond the first 64MB allocate additional * buffers to cover 1/10 of our ram over 64MB. When auto-sizing * the buffer cache we limit the eventual kva reservation to * maxbcache bytes. * * factor represents the 1/4 x ram conversion. */ if (nbuf == 0) { int factor = 4 * BKVASIZE / 1024; nbuf = 50; if (physmem_est > 4096) nbuf += min((physmem_est - 4096) / factor, 65536 / factor); if (physmem_est > 65536) nbuf += min((physmem_est - 65536) * 2 / (factor * 5), 32 * 1024 * 1024 / (factor * 5)); if (maxbcache && nbuf > maxbcache / BKVASIZE) nbuf = maxbcache / BKVASIZE; tuned_nbuf = 1; } else tuned_nbuf = 0; /* XXX Avoid unsigned long overflows later on with maxbufspace. */ maxbuf = (LONG_MAX / 3) / BKVASIZE; if (nbuf > maxbuf) { if (!tuned_nbuf) printf("Warning: nbufs lowered from %d to %ld\n", nbuf, maxbuf); nbuf = maxbuf; } /* * Ideal allocation size for the transient bio submap is 10% * of the maximal space buffer map. This roughly corresponds * to the amount of the buffer mapped for typical UFS load. * * Clip the buffer map to reserve space for the transient * BIOs, if its extent is bigger than 90% (80% on i386) of the * maximum buffer map extent on the platform. * * The fall-back to the maxbuf in case of maxbcache unset, * allows to not trim the buffer KVA for the architectures * with ample KVA space. */ if (bio_transient_maxcnt == 0 && unmapped_buf_allowed) { maxbuf_sz = maxbcache != 0 ? maxbcache : maxbuf * BKVASIZE; buf_sz = (long)nbuf * BKVASIZE; if (buf_sz < maxbuf_sz / TRANSIENT_DENOM * (TRANSIENT_DENOM - 1)) { /* * There is more KVA than memory. Do not * adjust buffer map size, and assign the rest * of maxbuf to transient map. */ biotmap_sz = maxbuf_sz - buf_sz; } else { /* * Buffer map spans all KVA we could afford on * this platform. Give 10% (20% on i386) of * the buffer map to the transient bio map. */ biotmap_sz = buf_sz / TRANSIENT_DENOM; buf_sz -= biotmap_sz; } if (biotmap_sz / INT_MAX > MAXPHYS) bio_transient_maxcnt = INT_MAX; else bio_transient_maxcnt = biotmap_sz / MAXPHYS; /* * Artificially limit to 1024 simultaneous in-flight I/Os * using the transient mapping. */ if (bio_transient_maxcnt > 1024) bio_transient_maxcnt = 1024; if (tuned_nbuf) nbuf = buf_sz / BKVASIZE; } /* * swbufs are used as temporary holders for I/O, such as paging I/O. * We have no less then 16 and no more then 256. */ nswbuf = min(nbuf / 4, 256); TUNABLE_INT_FETCH("kern.nswbuf", &nswbuf); if (nswbuf < NSWBUF_MIN) nswbuf = NSWBUF_MIN; /* * Reserve space for the buffer cache buffers */ swbuf = (void *)v; v = (caddr_t)(swbuf + nswbuf); buf = (void *)v; v = (caddr_t)(buf + nbuf); return(v); } /* Initialize the buffer subsystem. Called before use of any buffers. */ void bufinit(void) { struct buf *bp; int i; - CTASSERT(MAXBCACHEBUF >= MAXBSIZE); + KASSERT(maxbcachebuf >= MAXBSIZE, + ("maxbcachebuf (%d) must be >= MAXBSIZE (%d)\n", maxbcachebuf, + MAXBSIZE)); mtx_init(&bqlocks[QUEUE_DIRTY], "bufq dirty lock", NULL, MTX_DEF); mtx_init(&bqlocks[QUEUE_EMPTY], "bufq empty lock", NULL, MTX_DEF); for (i = QUEUE_CLEAN; i < QUEUE_CLEAN + CLEAN_QUEUES; i++) mtx_init(&bqlocks[i], "bufq clean lock", NULL, MTX_DEF); mtx_init(&rbreqlock, "runningbufspace lock", NULL, MTX_DEF); rw_init(&nblock, "needsbuffer lock"); mtx_init(&bdlock, "buffer daemon lock", NULL, MTX_DEF); mtx_init(&bdirtylock, "dirty buf lock", NULL, MTX_DEF); /* next, make a null set of free lists */ for (i = 0; i < BUFFER_QUEUES; i++) TAILQ_INIT(&bufqueues[i]); unmapped_buf = (caddr_t)kva_alloc(MAXPHYS); /* finally, initialize each buffer header and stick on empty q */ for (i = 0; i < nbuf; i++) { bp = &buf[i]; bzero(bp, sizeof *bp); bp->b_flags = B_INVAL; bp->b_rcred = NOCRED; bp->b_wcred = NOCRED; bp->b_qindex = QUEUE_EMPTY; bp->b_xflags = 0; bp->b_data = bp->b_kvabase = unmapped_buf; LIST_INIT(&bp->b_dep); BUF_LOCKINIT(bp); TAILQ_INSERT_TAIL(&bufqueues[QUEUE_EMPTY], bp, b_freelist); #ifdef INVARIANTS bq_len[QUEUE_EMPTY]++; #endif } /* * maxbufspace is the absolute maximum amount of buffer space we are * allowed to reserve in KVM and in real terms. The absolute maximum * is nominally used by metadata. hibufspace is the nominal maximum * used by most other requests. The differential is required to * ensure that metadata deadlocks don't occur. * * maxbufspace is based on BKVASIZE. Allocating buffers larger then * this may result in KVM fragmentation which is not handled optimally * by the system. XXX This is less true with vmem. We could use * PAGE_SIZE. */ maxbufspace = (long)nbuf * BKVASIZE; - hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - MAXBCACHEBUF * 10); + hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - maxbcachebuf * 10); lobufspace = (hibufspace / 20) * 19; /* 95% */ bufspacethresh = lobufspace + (hibufspace - lobufspace) / 2; /* * Note: The 16 MiB upper limit for hirunningspace was chosen * arbitrarily and may need further tuning. It corresponds to * 128 outstanding write IO requests (if IO size is 128 KiB), * which fits with many RAID controllers' tagged queuing limits. * The lower 1 MiB limit is the historical upper limit for * hirunningspace. */ - hirunningspace = lmax(lmin(roundup(hibufspace / 64, MAXBCACHEBUF), + hirunningspace = lmax(lmin(roundup(hibufspace / 64, maxbcachebuf), 16 * 1024 * 1024), 1024 * 1024); - lorunningspace = roundup((hirunningspace * 2) / 3, MAXBCACHEBUF); + lorunningspace = roundup((hirunningspace * 2) / 3, maxbcachebuf); /* * Limit the amount of malloc memory since it is wired permanently into * the kernel space. Even though this is accounted for in the buffer * allocation, we don't want the malloced region to grow uncontrolled. * The malloc scheme improves memory utilization significantly on * average (small) directories. */ maxbufmallocspace = hibufspace / 20; /* * Reduce the chance of a deadlock occurring by limiting the number * of delayed-write dirty buffers we allow to stack up. */ hidirtybuffers = nbuf / 4 + 20; dirtybufthresh = hidirtybuffers * 9 / 10; numdirtybuffers = 0; /* * To support extreme low-memory systems, make sure hidirtybuffers * cannot eat up all available buffer space. This occurs when our * minimum cannot be met. We try to size hidirtybuffers to 3/4 our * buffer space assuming BKVASIZE'd buffers. */ while ((long)hidirtybuffers * BKVASIZE > 3 * hibufspace / 4) { hidirtybuffers >>= 1; } lodirtybuffers = hidirtybuffers / 2; /* * lofreebuffers should be sufficient to avoid stalling waiting on * buf headers under heavy utilization. The bufs in per-cpu caches * are counted as free but will be unavailable to threads executing * on other cpus. * * hifreebuffers is the free target for the bufspace daemon. This * should be set appropriately to limit work per-iteration. */ lofreebuffers = MIN((nbuf / 25) + (20 * mp_ncpus), 128 * mp_ncpus); hifreebuffers = (3 * lofreebuffers) / 2; numfreebuffers = nbuf; /* Setup the kva and free list allocators. */ vmem_set_reclaim(buffer_arena, bufkva_reclaim); buf_zone = uma_zcache_create("buf free cache", sizeof(struct buf), NULL, NULL, NULL, NULL, buf_import, buf_release, NULL, 0); /* * Size the clean queue according to the amount of buffer space. * One queue per-256mb up to the max. More queues gives better * concurrency but less accurate LRU. */ clean_queues = MIN(howmany(maxbufspace, 256*1024*1024), CLEAN_QUEUES); } #ifdef INVARIANTS static inline void vfs_buf_check_mapped(struct buf *bp) { KASSERT(bp->b_kvabase != unmapped_buf, ("mapped buf: b_kvabase was not updated %p", bp)); KASSERT(bp->b_data != unmapped_buf, ("mapped buf: b_data was not updated %p", bp)); KASSERT(bp->b_data < unmapped_buf || bp->b_data >= unmapped_buf + MAXPHYS, ("b_data + b_offset unmapped %p", bp)); } static inline void vfs_buf_check_unmapped(struct buf *bp) { KASSERT(bp->b_data == unmapped_buf, ("unmapped buf: corrupted b_data %p", bp)); } #define BUF_CHECK_MAPPED(bp) vfs_buf_check_mapped(bp) #define BUF_CHECK_UNMAPPED(bp) vfs_buf_check_unmapped(bp) #else #define BUF_CHECK_MAPPED(bp) do {} while (0) #define BUF_CHECK_UNMAPPED(bp) do {} while (0) #endif static int isbufbusy(struct buf *bp) { if (((bp->b_flags & B_INVAL) == 0 && BUF_ISLOCKED(bp)) || ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)) return (1); return (0); } /* * Shutdown the system cleanly to prepare for reboot, halt, or power off. */ void bufshutdown(int show_busybufs) { static int first_buf_printf = 1; struct buf *bp; int iter, nbusy, pbusy; #ifndef PREEMPTION int subiter; #endif /* * Sync filesystems for shutdown */ wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); /* * With soft updates, some buffers that are * written will be remarked as dirty until other * buffers are written. */ for (iter = pbusy = 0; iter < 20; iter++) { nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) if (isbufbusy(bp)) nbusy++; if (nbusy == 0) { if (first_buf_printf) printf("All buffers synced."); break; } if (first_buf_printf) { printf("Syncing disks, buffers remaining... "); first_buf_printf = 0; } printf("%d ", nbusy); if (nbusy < pbusy) iter = 0; pbusy = nbusy; wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); #ifdef PREEMPTION /* * Drop Giant and spin for a while to allow * interrupt threads to run. */ DROP_GIANT(); DELAY(50000 * iter); PICKUP_GIANT(); #else /* * Drop Giant and context switch several times to * allow interrupt threads to run. */ DROP_GIANT(); for (subiter = 0; subiter < 50 * iter; subiter++) { thread_lock(curthread); mi_switch(SW_VOL, NULL); thread_unlock(curthread); DELAY(1000); } PICKUP_GIANT(); #endif } printf("\n"); /* * Count only busy local buffers to prevent forcing * a fsck if we're just a client of a wedged NFS server */ nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) { if (isbufbusy(bp)) { #if 0 /* XXX: This is bogus. We should probably have a BO_REMOTE flag instead */ if (bp->b_dev == NULL) { TAILQ_REMOVE(&mountlist, bp->b_vp->v_mount, mnt_list); continue; } #endif nbusy++; if (show_busybufs > 0) { printf( "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:", nbusy, bp, bp->b_vp, bp->b_flags, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); BUF_LOCKPRINTINFO(bp); if (show_busybufs > 1) vn_printf(bp->b_vp, "vnode content: "); } } } if (nbusy) { /* * Failed to sync all blocks. Indicate this and don't * unmount filesystems (thus forcing an fsck on reboot). */ printf("Giving up on %d buffers\n", nbusy); DELAY(5000000); /* 5 seconds */ } else { if (!first_buf_printf) printf("Final sync complete\n"); /* * Unmount filesystems */ if (panicstr == NULL) vfs_unmountall(); } swapoff_all(); DELAY(100000); /* wait for console output to finish */ } static void bpmap_qenter(struct buf *bp) { BUF_CHECK_MAPPED(bp); /* * bp->b_data is relative to bp->b_offset, but * bp->b_offset may be offset into the first page. */ bp->b_data = (caddr_t)trunc_page((vm_offset_t)bp->b_data); pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); bp->b_data = (caddr_t)((vm_offset_t)bp->b_data | (vm_offset_t)(bp->b_offset & PAGE_MASK)); } /* * binsfree: * * Insert the buffer into the appropriate free list. */ static void binsfree(struct buf *bp, int qindex) { struct mtx *olock, *nlock; if (qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } /* * Stick to the same clean queue for the lifetime of the buf to * limit locking below. Otherwise pick ont sequentially. */ if (qindex == QUEUE_CLEAN) { if (bqisclean(bp->b_qindex)) qindex = bp->b_qindex; else qindex = bqcleanq(); } /* * Handle delayed bremfree() processing. */ nlock = bqlock(qindex); if (bp->b_flags & B_REMFREE) { olock = bqlock(bp->b_qindex); mtx_lock(olock); bremfreel(bp); if (olock != nlock) { mtx_unlock(olock); mtx_lock(nlock); } } else mtx_lock(nlock); if (bp->b_qindex != QUEUE_NONE) panic("binsfree: free buffer onto another queue???"); bp->b_qindex = qindex; if (bp->b_flags & B_AGE) TAILQ_INSERT_HEAD(&bufqueues[bp->b_qindex], bp, b_freelist); else TAILQ_INSERT_TAIL(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS bq_len[bp->b_qindex]++; #endif mtx_unlock(nlock); } /* * buf_free: * * Free a buffer to the buf zone once it no longer has valid contents. */ static void buf_free(struct buf *bp) { if (bp->b_flags & B_REMFREE) bremfreef(bp); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 1"); if (bp->b_rcred != NOCRED) { crfree(bp->b_rcred); bp->b_rcred = NOCRED; } if (bp->b_wcred != NOCRED) { crfree(bp->b_wcred); bp->b_wcred = NOCRED; } if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); bufkva_free(bp); BUF_UNLOCK(bp); uma_zfree(buf_zone, bp); atomic_add_int(&numfreebuffers, 1); bufspace_wakeup(); } /* * buf_import: * * Import bufs into the uma cache from the buf list. The system still * expects a static array of bufs and much of the synchronization * around bufs assumes type stable storage. As a result, UMA is used * only as a per-cpu cache of bufs still maintained on a global list. */ static int buf_import(void *arg, void **store, int cnt, int flags) { struct buf *bp; int i; mtx_lock(&bqlocks[QUEUE_EMPTY]); for (i = 0; i < cnt; i++) { bp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTY]); if (bp == NULL) break; bremfreel(bp); store[i] = bp; } mtx_unlock(&bqlocks[QUEUE_EMPTY]); return (i); } /* * buf_release: * * Release bufs from the uma cache back to the buffer queues. */ static void buf_release(void *arg, void **store, int cnt) { int i; for (i = 0; i < cnt; i++) binsfree(store[i], QUEUE_EMPTY); } /* * buf_alloc: * * Allocate an empty buffer header. */ static struct buf * buf_alloc(void) { struct buf *bp; bp = uma_zalloc(buf_zone, M_NOWAIT); if (bp == NULL) { bufspace_daemonwakeup(); atomic_add_int(&numbufallocfails, 1); return (NULL); } /* * Wake-up the bufspace daemon on transition. */ if (atomic_fetchadd_int(&numfreebuffers, -1) == lofreebuffers) bufspace_daemonwakeup(); if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) panic("getnewbuf_empty: Locked buf %p on free queue.", bp); KASSERT(bp->b_vp == NULL, ("bp: %p still has vnode %p.", bp, bp->b_vp)); KASSERT((bp->b_flags & (B_DELWRI | B_NOREUSE)) == 0, ("invalid buffer %p flags %#x", bp, bp->b_flags)); KASSERT((bp->b_xflags & (BX_VNCLEAN|BX_VNDIRTY)) == 0, ("bp: %p still on a buffer list. xflags %X", bp, bp->b_xflags)); KASSERT(bp->b_npages == 0, ("bp: %p still has %d vm pages\n", bp, bp->b_npages)); KASSERT(bp->b_kvasize == 0, ("bp: %p still has kva\n", bp)); KASSERT(bp->b_bufsize == 0, ("bp: %p still has bufspace\n", bp)); bp->b_flags = 0; bp->b_ioflags = 0; bp->b_xflags = 0; bp->b_vflags = 0; bp->b_vp = NULL; bp->b_blkno = bp->b_lblkno = 0; bp->b_offset = NOOFFSET; bp->b_iodone = 0; bp->b_error = 0; bp->b_resid = 0; bp->b_bcount = 0; bp->b_npages = 0; bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_bufobj = NULL; bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_fsprivate1 = NULL; bp->b_fsprivate2 = NULL; bp->b_fsprivate3 = NULL; LIST_INIT(&bp->b_dep); return (bp); } /* * buf_qrecycle: * * Free a buffer from the given bufqueue. kva controls whether the * freed buf must own some kva resources. This is used for * defragmenting. */ static int buf_qrecycle(int qindex, bool kva) { struct buf *bp, *nbp; if (kva) atomic_add_int(&bufdefragcnt, 1); nbp = NULL; mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); /* * Run scan, possibly freeing data and/or kva mappings on the fly * depending. */ while ((bp = nbp) != NULL) { /* * Calculate next bp (we can only use it if we do not * release the bqlock). */ nbp = TAILQ_NEXT(bp, b_freelist); /* * If we are defragging then we need a buffer with * some kva to reclaim. */ if (kva && bp->b_kvasize == 0) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) continue; /* * Skip buffers with background writes in progress. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0) { BUF_UNLOCK(bp); continue; } KASSERT(bp->b_qindex == qindex, ("getnewbuf: inconsistent queue %d bp %p", qindex, bp)); /* * NOTE: nbp is now entirely invalid. We can only restart * the scan from this point on. */ bremfreel(bp); mtx_unlock(&bqlocks[qindex]); /* * Requeue the background write buffer with error and * restart the scan. */ if ((bp->b_vflags & BV_BKGRDERR) != 0) { bqrelse(bp); mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); continue; } bp->b_flags |= B_INVAL; brelse(bp); return (0); } mtx_unlock(&bqlocks[qindex]); return (ENOBUFS); } /* * buf_recycle: * * Iterate through all clean queues until we find a buf to recycle or * exhaust the search. */ static int buf_recycle(bool kva) { int qindex, first_qindex; qindex = first_qindex = bqcleanq(); do { if (buf_qrecycle(qindex, kva) == 0) return (0); if (++qindex == QUEUE_CLEAN + clean_queues) qindex = QUEUE_CLEAN; } while (qindex != first_qindex); return (ENOBUFS); } /* * buf_scan: * * Scan the clean queues looking for a buffer to recycle. needsbuffer * is set on failure so that the caller may optionally bufspace_wait() * in a race-free fashion. */ static int buf_scan(bool defrag) { int error; /* * To avoid heavy synchronization and wakeup races we set * needsbuffer and re-poll before failing. This ensures that * no frees can be missed between an unsuccessful poll and * going to sleep in a synchronized fashion. */ if ((error = buf_recycle(defrag)) != 0) { atomic_set_int(&needsbuffer, 1); bufspace_daemonwakeup(); error = buf_recycle(defrag); } if (error == 0) atomic_add_int(&getnewbufrestarts, 1); return (error); } /* * bremfree: * * Mark the buffer for removal from the appropriate free list. * */ void bremfree(struct buf *bp) { CTR3(KTR_BUF, "bremfree(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT((bp->b_flags & B_REMFREE) == 0, ("bremfree: buffer %p already marked for delayed removal.", bp)); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfree: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); bp->b_flags |= B_REMFREE; } /* * bremfreef: * * Force an immediate removal from a free list. Used only in nfs when * it abuses the b_freelist pointer. */ void bremfreef(struct buf *bp) { struct mtx *qlock; qlock = bqlock(bp->b_qindex); mtx_lock(qlock); bremfreel(bp); mtx_unlock(qlock); } /* * bremfreel: * * Removes a buffer from the free list, must be called with the * correct qlock held. */ static void bremfreel(struct buf *bp) { CTR3(KTR_BUF, "bremfreel(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfreel: buffer %p not on a queue.", bp)); if (bp->b_qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } mtx_assert(bqlock(bp->b_qindex), MA_OWNED); TAILQ_REMOVE(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS KASSERT(bq_len[bp->b_qindex] >= 1, ("queue %d underflow", bp->b_qindex)); bq_len[bp->b_qindex]--; #endif bp->b_qindex = QUEUE_NONE; bp->b_flags &= ~B_REMFREE; } /* * bufkva_free: * * Free the kva allocation for a buffer. * */ static void bufkva_free(struct buf *bp) { #ifdef INVARIANTS if (bp->b_kvasize == 0) { KASSERT(bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf, ("Leaked KVA space on %p", bp)); } else if (buf_mapped(bp)) BUF_CHECK_MAPPED(bp); else BUF_CHECK_UNMAPPED(bp); #endif if (bp->b_kvasize == 0) return; vmem_free(buffer_arena, (vm_offset_t)bp->b_kvabase, bp->b_kvasize); atomic_subtract_long(&bufkvaspace, bp->b_kvasize); atomic_add_int(&buffreekvacnt, 1); bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_kvasize = 0; } /* * bufkva_alloc: * * Allocate the buffer KVA and set b_kvasize and b_kvabase. */ static int bufkva_alloc(struct buf *bp, int maxsize, int gbflags) { vm_offset_t addr; int error; KASSERT((gbflags & GB_UNMAPPED) == 0 || (gbflags & GB_KVAALLOC) != 0, ("Invalid gbflags 0x%x in %s", gbflags, __func__)); bufkva_free(bp); addr = 0; error = vmem_alloc(buffer_arena, maxsize, M_BESTFIT | M_NOWAIT, &addr); if (error != 0) { /* * Buffer map is too fragmented. Request the caller * to defragment the map. */ return (error); } bp->b_kvabase = (caddr_t)addr; bp->b_kvasize = maxsize; atomic_add_long(&bufkvaspace, bp->b_kvasize); if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; BUF_CHECK_UNMAPPED(bp); } else { bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); } return (0); } /* * bufkva_reclaim: * * Reclaim buffer kva by freeing buffers holding kva. This is a vmem * callback that fires to avoid returning failure. */ static void bufkva_reclaim(vmem_t *vmem, int flags) { int i; for (i = 0; i < 5; i++) if (buf_scan(true) != 0) break; return; } /* * Attempt to initiate asynchronous I/O on read-ahead blocks. We must * clear BIO_ERROR and B_INVAL prior to initiating I/O . If B_CACHE is set, * the buffer is valid and we do not have to do anything. */ void breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt, struct ucred * cred) { struct buf *rabp; int i; for (i = 0; i < cnt; i++, rablkno++, rabsize++) { if (inmem(vp, *rablkno)) continue; rabp = getblk(vp, *rablkno, *rabsize, 0, 0, 0); if ((rabp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, rabp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } rabp->b_flags |= B_ASYNC; rabp->b_flags &= ~B_INVAL; rabp->b_ioflags &= ~BIO_ERROR; rabp->b_iocmd = BIO_READ; if (rabp->b_rcred == NOCRED && cred != NOCRED) rabp->b_rcred = crhold(cred); vfs_busy_pages(rabp, 0); BUF_KERNPROC(rabp); rabp->b_iooffset = dbtob(rabp->b_blkno); bstrategy(rabp); } else { brelse(rabp); } } } /* * Entry point for bread() and breadn() via #defines in sys/buf.h. * * Get a buffer with the specified data. Look in the cache first. We * must clear BIO_ERROR and B_INVAL prior to initiating I/O. If B_CACHE * is set, the buffer is valid and we do not have to do anything, see * getblk(). Also starts asynchronous I/O on read-ahead blocks. * * Always return a NULL buffer pointer (in bpp) when returning an error. */ int breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno, int *rabsize, int cnt, struct ucred *cred, int flags, struct buf **bpp) { struct buf *bp; int rv = 0, readwait = 0; CTR3(KTR_BUF, "breadn(%p, %jd, %d)", vp, blkno, size); /* * Can only return NULL if GB_LOCK_NOWAIT flag is specified. */ *bpp = bp = getblk(vp, blkno, size, 0, 0, flags); if (bp == NULL) return (EBUSY); /* if not found in cache, do some I/O */ if ((bp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if (bp->b_rcred == NOCRED && cred != NOCRED) bp->b_rcred = crhold(cred); vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); ++readwait; } breada(vp, rablkno, rabsize, cnt, cred); if (readwait) { rv = bufwait(bp); if (rv != 0) { brelse(bp); *bpp = NULL; } } return (rv); } /* * Write, release buffer on completion. (Done by iodone * if async). Do not bother writing anything if the buffer * is invalid. * * Note that we set B_CACHE here, indicating that buffer is * fully valid and thus cacheable. This is true even of NFS * now so we set it generally. This could be set either here * or in biodone() since the I/O is synchronous. We put it * here. */ int bufwrite(struct buf *bp) { int oldflags; struct vnode *vp; long space; int vp_md; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if ((bp->b_bufobj->bo_flag & BO_DEAD) != 0) { bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_CACHE; brelse(bp); return (ENXIO); } if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (bp->b_flags & B_BARRIER) barrierwrites++; oldflags = bp->b_flags; BUF_ASSERT_HELD(bp); KASSERT(!(bp->b_vflags & BV_BKGRDINPROG), ("FFS background buffer should not get here %p", bp)); vp = bp->b_vp; if (vp) vp_md = vp->v_vflag & VV_MD; else vp_md = 0; /* * Mark the buffer clean. Increment the bufobj write count * before bundirty() call, to prevent other thread from seeing * empty dirty list and zero counter for writes in progress, * falsely indicating that the bufobj is clean. */ bufobj_wref(bp->b_bufobj); bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_flags |= B_CACHE; bp->b_iocmd = BIO_WRITE; vfs_busy_pages(bp, 1); /* * Normal bwrites pipeline writes */ bp->b_runningbufspace = bp->b_bufsize; space = atomic_fetchadd_long(&runningbufspace, bp->b_runningbufspace); if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_oublock++; } if (oldflags & B_ASYNC) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); buf_track(bp, __func__); bstrategy(bp); if ((oldflags & B_ASYNC) == 0) { int rtval = bufwait(bp); brelse(bp); return (rtval); } else if (space > hirunningspace) { /* * don't allow the async write to saturate the I/O * system. We will not deadlock here because * we are blocking waiting for I/O that is already in-progress * to complete. We do not block here if it is the update * or syncer daemon trying to clean up as that can lead * to deadlock. */ if ((curthread->td_pflags & TDP_NORUNNINGBUF) == 0 && !vp_md) waitrunningbufspace(); } return (0); } void bufbdflush(struct bufobj *bo, struct buf *bp) { struct buf *nbp; if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10) { (void) VOP_FSYNC(bp->b_vp, MNT_NOWAIT, curthread); altbufferflushes++; } else if (bo->bo_dirty.bv_cnt > dirtybufthresh) { BO_LOCK(bo); /* * Try to find a buffer to flush. */ TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { if ((nbp->b_vflags & BV_BKGRDINPROG) || BUF_LOCK(nbp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if (bp == nbp) panic("bdwrite: found ourselves"); BO_UNLOCK(bo); /* Don't countdeps with the bo lock held. */ if (buf_countdeps(nbp, 0)) { BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } if (nbp->b_flags & B_CLUSTEROK) { vfs_bio_awrite(nbp); } else { bremfree(nbp); bawrite(nbp); } dirtybufferflushes++; break; } if (nbp == NULL) BO_UNLOCK(bo); } } /* * Delayed write. (Buffer is marked dirty). Do not bother writing * anything if the buffer is marked invalid. * * Note that since the buffer must be completely valid, we can safely * set B_CACHE. In fact, we have to set B_CACHE here rather then in * biodone() in order to prevent getblk from writing the buffer * out synchronously. */ void bdwrite(struct buf *bp) { struct thread *td = curthread; struct vnode *vp; struct bufobj *bo; CTR3(KTR_BUF, "bdwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT((bp->b_flags & B_BARRIER) == 0, ("Barrier request in delayed write %p", bp)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_INVAL) { brelse(bp); return; } /* * If we have too many dirty buffers, don't create any more. * If we are wildly over our limit, then force a complete * cleanup. Otherwise, just keep the situation from getting * out of control. Note that we have to avoid a recursive * disaster and not try to clean up after our own cleanup! */ vp = bp->b_vp; bo = bp->b_bufobj; if ((td->td_pflags & (TDP_COWINPROGRESS|TDP_INBDFLUSH)) == 0) { td->td_pflags |= TDP_INBDFLUSH; BO_BDFLUSH(bo, bp); td->td_pflags &= ~TDP_INBDFLUSH; } else recursiveflushes++; bdirty(bp); /* * Set B_CACHE, indicating that the buffer is fully valid. This is * true even of NFS now. */ bp->b_flags |= B_CACHE; /* * This bmap keeps the system from needing to do the bmap later, * perhaps when the system is attempting to do a sync. Since it * is likely that the indirect block -- or whatever other datastructure * that the filesystem needs is still in memory now, it is a good * thing to do this. Note also, that if the pageout daemon is * requesting a sync -- there might not be enough memory to do * the bmap then... So, this is important to do. */ if (vp->v_type != VCHR && bp->b_lblkno == bp->b_blkno) { VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } buf_track(bp, __func__); /* * Set the *dirty* buffer range based upon the VM system dirty * pages. * * Mark the buffer pages as clean. We need to do this here to * satisfy the vnode_pager and the pageout daemon, so that it * thinks that the pages have been "cleaned". Note that since * the pages are in a delayed write buffer -- the VFS layer * "will" see that the pages get written out on the next sync, * or perhaps the cluster will be completed. */ vfs_clean_pages_dirty_buf(bp); bqrelse(bp); /* * note: we cannot initiate I/O from a bdwrite even if we wanted to, * due to the softdep code. */ } /* * bdirty: * * Turn buffer into delayed write request. We must clear BIO_READ and * B_RELBUF, and we must set B_DELWRI. We reassign the buffer to * itself to properly update it in the dirty/clean lists. We mark it * B_DONE to ensure that any asynchronization of the buffer properly * clears B_DONE ( else a panic will occur later ). * * bdirty() is kinda like bdwrite() - we have to clear B_INVAL which * might have been set pre-getblk(). Unlike bwrite/bdwrite, bdirty() * should only be called if the buffer is known-good. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bdirty(struct buf *bp) { CTR3(KTR_BUF, "bdirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bdirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); bp->b_flags &= ~(B_RELBUF); bp->b_iocmd = BIO_WRITE; if ((bp->b_flags & B_DELWRI) == 0) { bp->b_flags |= /* XXX B_DONE | */ B_DELWRI; reassignbuf(bp); bdirtyadd(); } } /* * bundirty: * * Clear B_DELWRI for buffer. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bundirty(struct buf *bp) { CTR3(KTR_BUF, "bundirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bundirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_DELWRI) { bp->b_flags &= ~B_DELWRI; reassignbuf(bp); bdirtysub(); } /* * Since it is now being written, we can clear its deferred write flag. */ bp->b_flags &= ~B_DEFERRED; } /* * bawrite: * * Asynchronous write. Start output on a buffer, but do not wait for * it to complete. The buffer is released when the output completes. * * bwrite() ( or the VOP routine anyway ) is responsible for handling * B_INVAL buffers. Not us. */ void bawrite(struct buf *bp) { bp->b_flags |= B_ASYNC; (void) bwrite(bp); } /* * babarrierwrite: * * Asynchronous barrier write. Start output on a buffer, but do not * wait for it to complete. Place a write barrier after this write so * that this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ void babarrierwrite(struct buf *bp) { bp->b_flags |= B_ASYNC | B_BARRIER; (void) bwrite(bp); } /* * bbarrierwrite: * * Synchronous barrier write. Start output on a buffer and wait for * it to complete. Place a write barrier after this write so that * this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ int bbarrierwrite(struct buf *bp) { bp->b_flags |= B_BARRIER; return (bwrite(bp)); } /* * bwillwrite: * * Called prior to the locking of any vnodes when we are expecting to * write. We do not want to starve the buffer cache with too many * dirty buffers so we block here. By blocking prior to the locking * of any vnodes we attempt to avoid the situation where a locked vnode * prevents the various system daemons from flushing related buffers. */ void bwillwrite(void) { if (numdirtybuffers >= hidirtybuffers) { mtx_lock(&bdirtylock); while (numdirtybuffers >= hidirtybuffers) { bdirtywait = 1; msleep(&bdirtywait, &bdirtylock, (PRIBIO + 4), "flswai", 0); } mtx_unlock(&bdirtylock); } } /* * Return true if we have too many dirty buffers. */ int buf_dirty_count_severe(void) { return(numdirtybuffers >= hidirtybuffers); } /* * brelse: * * Release a busy buffer and, if requested, free its resources. The * buffer will be stashed in the appropriate bufqueue[] allowing it * to be accessed later as a cache entity or reused for other purposes. */ void brelse(struct buf *bp) { int qindex; /* * Many functions erroneously call brelse with a NULL bp under rare * error conditions. Simply return when called with a NULL bp. */ if (bp == NULL) return; CTR3(KTR_BUF, "brelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); KASSERT((bp->b_flags & B_VMIO) != 0 || (bp->b_flags & B_NOREUSE) == 0, ("brelse: non-VMIO buffer marked NOREUSE")); if (BUF_LOCKRECURSED(bp)) { /* * Do not process, in particular, do not handle the * B_INVAL/B_RELBUF and do not release to free list. */ BUF_UNLOCK(bp); return; } if (bp->b_flags & B_MANAGED) { bqrelse(bp); return; } if ((bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); bdirty(bp); } if (bp->b_iocmd == BIO_WRITE && (bp->b_ioflags & BIO_ERROR) && (bp->b_error != ENXIO || !LIST_EMPTY(&bp->b_dep)) && !(bp->b_flags & B_INVAL)) { /* * Failed write, redirty. All errors except ENXIO (which * means the device is gone) are expected to be potentially * transient - underlying media might work if tried again * after EIO, and memory might be available after an ENOMEM. * * Do this also for buffers that failed with ENXIO, but have * non-empty dependencies - the soft updates code might need * to access the buffer to untangle them. * * Must clear BIO_ERROR to prevent pages from being scrapped. */ bp->b_ioflags &= ~BIO_ERROR; bdirty(bp); } else if ((bp->b_flags & (B_NOCACHE | B_INVAL)) || (bp->b_ioflags & BIO_ERROR) || (bp->b_bufsize <= 0)) { /* * Either a failed read I/O, or we were asked to free or not * cache the buffer, or we failed to write to a device that's * no longer present. */ bp->b_flags |= B_INVAL; if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_flags & B_DELWRI) bdirtysub(); bp->b_flags &= ~(B_DELWRI | B_CACHE); if ((bp->b_flags & B_VMIO) == 0) { allocbuf(bp, 0); if (bp->b_vp) brelvp(bp); } } /* * We must clear B_RELBUF if B_DELWRI is set. If vfs_vmio_truncate() * is called with B_DELWRI set, the underlying pages may wind up * getting freed causing a previous write (bdwrite()) to get 'lost' * because pages associated with a B_DELWRI bp are marked clean. * * We still allow the B_INVAL case to call vfs_vmio_truncate(), even * if B_DELWRI is set. */ if (bp->b_flags & B_DELWRI) bp->b_flags &= ~B_RELBUF; /* * VMIO buffer rundown. It is not very necessary to keep a VMIO buffer * constituted, not even NFS buffers now. Two flags effect this. If * B_INVAL, the struct buf is invalidated but the VM object is kept * around ( i.e. so it is trivial to reconstitute the buffer later ). * * If BIO_ERROR or B_NOCACHE is set, pages in the VM object will be * invalidated. BIO_ERROR cannot be set for a failed write unless the * buffer is also B_INVAL because it hits the re-dirtying code above. * * Normally we can do this whether a buffer is B_DELWRI or not. If * the buffer is an NFS buffer, it is tracking piecemeal writes or * the commit state and we cannot afford to lose the buffer. If the * buffer has a background write in progress, we need to keep it * around to prevent it from being reconstituted and starting a second * background write. */ if ((bp->b_flags & B_VMIO) && (bp->b_flags & B_NOCACHE || (bp->b_ioflags & BIO_ERROR && bp->b_iocmd == BIO_READ)) && !(bp->b_vp->v_mount != NULL && (bp->b_vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && !vn_isdisk(bp->b_vp, NULL) && (bp->b_flags & B_DELWRI))) { vfs_vmio_invalidate(bp); allocbuf(bp, 0); } if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0 || (bp->b_flags & (B_DELWRI | B_NOREUSE)) == B_NOREUSE) { allocbuf(bp, 0); bp->b_flags &= ~B_NOREUSE; if (bp->b_vp != NULL) brelvp(bp); } /* * If the buffer has junk contents signal it and eventually * clean up B_DELWRI and diassociate the vnode so that gbincore() * doesn't find it. */ if (bp->b_bufsize == 0 || (bp->b_ioflags & BIO_ERROR) != 0 || (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) != 0) bp->b_flags |= B_INVAL; if (bp->b_flags & B_INVAL) { if (bp->b_flags & B_DELWRI) bundirty(bp); if (bp->b_vp) brelvp(bp); } buf_track(bp, __func__); /* buffers with no memory */ if (bp->b_bufsize == 0) { buf_free(bp); return; } /* buffers with junk contents */ if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF) || (bp->b_ioflags & BIO_ERROR)) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 2"); qindex = QUEUE_CLEAN; bp->b_flags |= B_AGE; /* remaining buffers */ } else if (bp->b_flags & B_DELWRI) qindex = QUEUE_DIRTY; else qindex = QUEUE_CLEAN; binsfree(bp, qindex); bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF | B_DIRECT); if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("brelse: not dirty"); /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * Release a buffer back to the appropriate queue but do not try to free * it. The buffer is expected to be used again soon. * * bqrelse() is used by bdwrite() to requeue a delayed write, and used by * biodone() to requeue an async I/O on completion. It is also used when * known good buffers need to be requeued but we think we may need the data * again soon. * * XXX we should be able to leave the B_RELBUF hint set on completion. */ void bqrelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "bqrelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); qindex = QUEUE_NONE; if (BUF_LOCKRECURSED(bp)) { /* do not release to free list */ BUF_UNLOCK(bp); return; } bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF); if (bp->b_flags & B_MANAGED) { if (bp->b_flags & B_REMFREE) bremfreef(bp); goto out; } /* buffers with stale but valid contents */ if ((bp->b_flags & B_DELWRI) != 0 || (bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); qindex = QUEUE_DIRTY; } else { if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("bqrelse: not dirty"); if ((bp->b_flags & B_NOREUSE) != 0) { brelse(bp); return; } qindex = QUEUE_CLEAN; } binsfree(bp, qindex); out: buf_track(bp, __func__); /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * Complete I/O to a VMIO backed page. Validate the pages as appropriate, * restore bogus pages. */ static void vfs_vmio_iodone(struct buf *bp) { vm_ooffset_t foff; vm_page_t m; vm_object_t obj; struct vnode *vp; int i, iosize, resid; bool bogus; obj = bp->b_bufobj->bo_object; KASSERT(obj->paging_in_progress >= bp->b_npages, ("vfs_vmio_iodone: paging in progress(%d) < b_npages(%d)", obj->paging_in_progress, bp->b_npages)); vp = bp->b_vp; KASSERT(vp->v_holdcnt > 0, ("vfs_vmio_iodone: vnode %p has zero hold count", vp)); KASSERT(vp->v_object != NULL, ("vfs_vmio_iodone: vnode %p has no vm_object", vp)); foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_vmio_iodone: bp %p has no buffer offset", bp)); bogus = false; iosize = bp->b_bcount - bp->b_resid; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff; if (resid > iosize) resid = iosize; /* * cleanup bogus pages, restoring the originals */ m = bp->b_pages[i]; if (m == bogus_page) { bogus = true; m = vm_page_lookup(obj, OFF_TO_IDX(foff)); if (m == NULL) panic("biodone: page disappeared!"); bp->b_pages[i] = m; } else if ((bp->b_iocmd == BIO_READ) && resid > 0) { /* * In the write case, the valid and clean bits are * already changed correctly ( see bdwrite() ), so we * only need to do this here in the read case. */ KASSERT((m->dirty & vm_page_bits(foff & PAGE_MASK, resid)) == 0, ("vfs_vmio_iodone: page %p " "has unexpected dirty bits", m)); vfs_page_set_valid(bp, foff, m); } KASSERT(OFF_TO_IDX(foff) == m->pindex, ("vfs_vmio_iodone: foff(%jd)/pindex(%ju) mismatch", (intmax_t)foff, (uintmax_t)m->pindex)); vm_page_sunbusy(m); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; iosize -= resid; } vm_object_pip_wakeupn(obj, bp->b_npages); VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * Unwire a page held by a buf and place it on the appropriate vm queue. */ static void vfs_vmio_unwire(struct buf *bp, vm_page_t m) { bool freed; vm_page_lock(m); if (vm_page_unwire(m, PQ_NONE)) { /* * Determine if the page should be freed before adding * it to the inactive queue. */ if (m->valid == 0) { freed = !vm_page_busied(m); if (freed) vm_page_free(m); } else if ((bp->b_flags & B_DIRECT) != 0) freed = vm_page_try_to_free(m); else freed = false; if (!freed) { /* * If the page is unlikely to be reused, let the * VM know. Otherwise, maintain LRU page * ordering and put the page at the tail of the * inactive queue. */ if ((bp->b_flags & B_NOREUSE) != 0) vm_page_deactivate_noreuse(m); else vm_page_deactivate(m); } } vm_page_unlock(m); } /* * Perform page invalidation when a buffer is released. The fully invalid * pages will be reclaimed later in vfs_vmio_truncate(). */ static void vfs_vmio_invalidate(struct buf *bp) { vm_object_t obj; vm_page_t m; int i, resid, poffset, presid; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove(trunc_page((vm_offset_t)bp->b_data), bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); /* * Get the base offset and length of the buffer. Note that * in the VMIO case if the buffer block size is not * page-aligned then b_data pointer may not be page-aligned. * But our b_pages[] array *IS* page aligned. * * block sizes less then DEV_BSIZE (usually 512) are not * supported due to the page granularity bits (m->valid, * m->dirty, etc...). * * See man buf(9) for more information */ obj = bp->b_bufobj->bo_object; resid = bp->b_bufsize; poffset = bp->b_offset & PAGE_MASK; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) panic("vfs_vmio_invalidate: Unexpected bogus page."); bp->b_pages[i] = NULL; presid = resid > (PAGE_SIZE - poffset) ? (PAGE_SIZE - poffset) : resid; KASSERT(presid >= 0, ("brelse: extra page")); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, "mbncsh", true); VM_OBJECT_WLOCK(obj); } if (pmap_page_wired_mappings(m) == 0) vm_page_set_invalid(m, poffset, presid); vfs_vmio_unwire(bp, m); resid -= presid; poffset = 0; } VM_OBJECT_WUNLOCK(obj); bp->b_npages = 0; } /* * Page-granular truncation of an existing VMIO buffer. */ static void vfs_vmio_truncate(struct buf *bp, int desiredpages) { vm_object_t obj; vm_page_t m; int i; if (bp->b_npages == desiredpages) return; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove((vm_offset_t)trunc_page((vm_offset_t)bp->b_data) + (desiredpages << PAGE_SHIFT), bp->b_npages - desiredpages); } else BUF_CHECK_UNMAPPED(bp); obj = bp->b_bufobj->bo_object; if (obj != NULL) VM_OBJECT_WLOCK(obj); for (i = desiredpages; i < bp->b_npages; i++) { m = bp->b_pages[i]; KASSERT(m != bogus_page, ("allocbuf: bogus page found")); bp->b_pages[i] = NULL; vfs_vmio_unwire(bp, m); } if (obj != NULL) VM_OBJECT_WUNLOCK(obj); bp->b_npages = desiredpages; } /* * Byte granular extension of VMIO buffers. */ static void vfs_vmio_extend(struct buf *bp, int desiredpages, int size) { /* * We are growing the buffer, possibly in a * byte-granular fashion. */ vm_object_t obj; vm_offset_t toff; vm_offset_t tinc; vm_page_t m; /* * Step 1, bring in the VM pages from the object, allocating * them if necessary. We must clear B_CACHE if these pages * are not valid for the range covered by the buffer. */ obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); while (bp->b_npages < desiredpages) { /* * We must allocate system pages since blocking * here could interfere with paging I/O, no * matter which process we are. * * Only exclusive busy can be tested here. * Blocking on shared busy might lead to * deadlocks once allocbuf() is called after * pages are vfs_busy_pages(). */ m = vm_page_grab(obj, OFF_TO_IDX(bp->b_offset) + bp->b_npages, VM_ALLOC_NOBUSY | VM_ALLOC_SYSTEM | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | VM_ALLOC_COUNT(desiredpages - bp->b_npages)); if (m->valid == 0) bp->b_flags &= ~B_CACHE; bp->b_pages[bp->b_npages] = m; ++bp->b_npages; } /* * Step 2. We've loaded the pages into the buffer, * we have to figure out if we can still have B_CACHE * set. Note that B_CACHE is set according to the * byte-granular range ( bcount and size ), not the * aligned range ( newbsize ). * * The VM test is against m->valid, which is DEV_BSIZE * aligned. Needless to say, the validity of the data * needs to also be DEV_BSIZE aligned. Note that this * fails with NFS if the server or some other client * extends the file's EOF. If our buffer is resized, * B_CACHE may remain set! XXX */ toff = bp->b_bcount; tinc = PAGE_SIZE - ((bp->b_offset + toff) & PAGE_MASK); while ((bp->b_flags & B_CACHE) && toff < size) { vm_pindex_t pi; if (tinc > (size - toff)) tinc = size - toff; pi = ((bp->b_offset & PAGE_MASK) + toff) >> PAGE_SHIFT; m = bp->b_pages[pi]; vfs_buf_test_cache(bp, bp->b_offset, toff, tinc, m); toff += tinc; tinc = PAGE_SIZE; } VM_OBJECT_WUNLOCK(obj); /* * Step 3, fixup the KVA pmap. */ if (buf_mapped(bp)) bpmap_qenter(bp); else BUF_CHECK_UNMAPPED(bp); } /* * Check to see if a block at a particular lbn is available for a clustered * write. */ static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno) { struct buf *bpa; int match; match = 0; /* If the buf isn't in core skip it */ if ((bpa = gbincore(&vp->v_bufobj, lblkno)) == NULL) return (0); /* If the buf is busy we don't want to wait for it */ if (BUF_LOCK(bpa, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) return (0); /* Only cluster with valid clusterable delayed write buffers */ if ((bpa->b_flags & (B_DELWRI | B_CLUSTEROK | B_INVAL)) != (B_DELWRI | B_CLUSTEROK)) goto done; if (bpa->b_bufsize != size) goto done; /* * Check to see if it is in the expected place on disk and that the * block has been mapped. */ if ((bpa->b_blkno != bpa->b_lblkno) && (bpa->b_blkno == blkno)) match = 1; done: BUF_UNLOCK(bpa); return (match); } /* * vfs_bio_awrite: * * Implement clustered async writes for clearing out B_DELWRI buffers. * This is much better then the old way of writing only one buffer at * a time. Note that we may not be presented with the buffers in the * correct order, so we search for the cluster in both directions. */ int vfs_bio_awrite(struct buf *bp) { struct bufobj *bo; int i; int j; daddr_t lblkno = bp->b_lblkno; struct vnode *vp = bp->b_vp; int ncl; int nwritten; int size; int maxcl; int gbflags; bo = &vp->v_bufobj; gbflags = (bp->b_data == unmapped_buf) ? GB_UNMAPPED : 0; /* * right now we support clustered writing only to regular files. If * we find a clusterable block we could be in the middle of a cluster * rather then at the beginning. */ if ((vp->v_type == VREG) && (vp->v_mount != 0) && /* Only on nodes that have the size info */ (bp->b_flags & (B_CLUSTEROK | B_INVAL)) == B_CLUSTEROK) { size = vp->v_mount->mnt_stat.f_iosize; maxcl = MAXPHYS / size; BO_RLOCK(bo); for (i = 1; i < maxcl; i++) if (vfs_bio_clcheck(vp, size, lblkno + i, bp->b_blkno + ((i * size) >> DEV_BSHIFT)) == 0) break; for (j = 1; i + j <= maxcl && j <= lblkno; j++) if (vfs_bio_clcheck(vp, size, lblkno - j, bp->b_blkno - ((j * size) >> DEV_BSHIFT)) == 0) break; BO_RUNLOCK(bo); --j; ncl = i + j; /* * this is a possible cluster write */ if (ncl != 1) { BUF_UNLOCK(bp); nwritten = cluster_wbuild(vp, size, lblkno - j, ncl, gbflags); return (nwritten); } } bremfree(bp); bp->b_flags |= B_ASYNC; /* * default (old) behavior, writing out only one block * * XXX returns b_bufsize instead of b_bcount for nwritten? */ nwritten = bp->b_bufsize; (void) bwrite(bp); return (nwritten); } /* * getnewbuf_kva: * * Allocate KVA for an empty buf header according to gbflags. */ static int getnewbuf_kva(struct buf *bp, int gbflags, int maxsize) { if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_UNMAPPED) { /* * In order to keep fragmentation sane we only allocate kva * in BKVASIZE chunks. XXX with vmem we can do page size. */ maxsize = (maxsize + BKVAMASK) & ~BKVAMASK; if (maxsize != bp->b_kvasize && bufkva_alloc(bp, maxsize, gbflags)) return (ENOSPC); } return (0); } /* * getnewbuf: * * Find and initialize a new buffer header, freeing up existing buffers * in the bufqueues as necessary. The new buffer is returned locked. * * We block if: * We have insufficient buffer headers * We have insufficient buffer space * buffer_arena is too fragmented ( space reservation fails ) * If we have to flush dirty buffers ( but we try to avoid this ) * * The caller is responsible for releasing the reserved bufspace after * allocbuf() is called. */ static struct buf * getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int maxsize, int gbflags) { struct buf *bp; bool metadata, reserved; bp = NULL; KASSERT((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); if (!unmapped_buf_allowed) gbflags &= ~(GB_UNMAPPED | GB_KVAALLOC); if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 || vp->v_type == VCHR) metadata = true; else metadata = false; atomic_add_int(&getnewbufcalls, 1); reserved = false; do { if (reserved == false && bufspace_reserve(maxsize, metadata) != 0) continue; reserved = true; if ((bp = buf_alloc()) == NULL) continue; if (getnewbuf_kva(bp, gbflags, maxsize) == 0) return (bp); break; } while(buf_scan(false) == 0); if (reserved) atomic_subtract_long(&bufspace, maxsize); if (bp != NULL) { bp->b_flags |= B_INVAL; brelse(bp); } bufspace_wait(vp, gbflags, slpflag, slptimeo); return (NULL); } /* * buf_daemon: * * buffer flushing daemon. Buffers are normally flushed by the * update daemon but if it cannot keep up this process starts to * take the load in an attempt to prevent getnewbuf() from blocking. */ static struct kproc_desc buf_kp = { "bufdaemon", buf_daemon, &bufdaemonproc }; SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &buf_kp); static int buf_flush(struct vnode *vp, int target) { int flushed; flushed = flushbufqueues(vp, target, 0); if (flushed == 0) { /* * Could not find any buffers without rollback * dependencies, so just write the first one * in the hopes of eventually making progress. */ if (vp != NULL && target > 2) target /= 2; flushbufqueues(vp, target, 1); } return (flushed); } static void buf_daemon() { int lodirty; /* * This process needs to be suspended prior to shutdown sync. */ EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, bufdaemonproc, SHUTDOWN_PRI_LAST); /* * This process is allowed to take the buffer cache to the limit */ curthread->td_pflags |= TDP_NORUNNINGBUF | TDP_BUFNEED; mtx_lock(&bdlock); for (;;) { bd_request = 0; mtx_unlock(&bdlock); kproc_suspend_check(bufdaemonproc); lodirty = lodirtybuffers; if (bd_speedupreq) { lodirty = numdirtybuffers / 2; bd_speedupreq = 0; } /* * Do the flush. Limit the amount of in-transit I/O we * allow to build up, otherwise we would completely saturate * the I/O system. */ while (numdirtybuffers > lodirty) { if (buf_flush(NULL, numdirtybuffers - lodirty) == 0) break; kern_yield(PRI_USER); } /* * Only clear bd_request if we have reached our low water * mark. The buf_daemon normally waits 1 second and * then incrementally flushes any dirty buffers that have * built up, within reason. * * If we were unable to hit our low water mark and couldn't * find any flushable buffers, we sleep for a short period * to avoid endless loops on unlockable buffers. */ mtx_lock(&bdlock); if (numdirtybuffers <= lodirtybuffers) { /* * We reached our low water mark, reset the * request and sleep until we are needed again. * The sleep is just so the suspend code works. */ bd_request = 0; /* * Do an extra wakeup in case dirty threshold * changed via sysctl and the explicit transition * out of shortfall was missed. */ bdirtywakeup(); if (runningbufspace <= lorunningspace) runningwakeup(); msleep(&bd_request, &bdlock, PVM, "psleep", hz); } else { /* * We couldn't find any flushable dirty buffers but * still have too many dirty buffers, we * have to sleep and try again. (rare) */ msleep(&bd_request, &bdlock, PVM, "qsleep", hz / 10); } } } /* * flushbufqueues: * * Try to flush a buffer in the dirty queue. We must be careful to * free up B_INVAL buffers instead of write them, which NFS is * particularly sensitive to. */ static int flushwithdeps = 0; SYSCTL_INT(_vfs, OID_AUTO, flushwithdeps, CTLFLAG_RW, &flushwithdeps, 0, "Number of buffers flushed with dependecies that require rollbacks"); static int flushbufqueues(struct vnode *lvp, int target, int flushdeps) { struct buf *sentinel; struct vnode *vp; struct mount *mp; struct buf *bp; int hasdeps; int flushed; int queue; int error; bool unlock; flushed = 0; queue = QUEUE_DIRTY; bp = NULL; sentinel = malloc(sizeof(struct buf), M_TEMP, M_WAITOK | M_ZERO); sentinel->b_qindex = QUEUE_SENTINEL; mtx_lock(&bqlocks[queue]); TAILQ_INSERT_HEAD(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); while (flushed != target) { maybe_yield(); mtx_lock(&bqlocks[queue]); bp = TAILQ_NEXT(sentinel, b_freelist); if (bp != NULL) { TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); TAILQ_INSERT_AFTER(&bufqueues[queue], bp, sentinel, b_freelist); } else { mtx_unlock(&bqlocks[queue]); break; } /* * Skip sentinels inserted by other invocations of the * flushbufqueues(), taking care to not reorder them. * * Only flush the buffers that belong to the * vnode locked by the curthread. */ if (bp->b_qindex == QUEUE_SENTINEL || (lvp != NULL && bp->b_vp != lvp)) { mtx_unlock(&bqlocks[queue]); continue; } error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); mtx_unlock(&bqlocks[queue]); if (error != 0) continue; /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0 || (bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); continue; } if (bp->b_flags & B_INVAL) { bremfreef(bp); brelse(bp); flushed++; continue; } if (!LIST_EMPTY(&bp->b_dep) && buf_countdeps(bp, 0)) { if (flushdeps == 0) { BUF_UNLOCK(bp); continue; } hasdeps = 1; } else hasdeps = 0; /* * We must hold the lock on a vnode before writing * one of its buffers. Otherwise we may confuse, or * in the case of a snapshot vnode, deadlock the * system. * * The lock order here is the reverse of the normal * of vnode followed by buf lock. This is ok because * the NOWAIT will prevent deadlock. */ vp = bp->b_vp; if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { BUF_UNLOCK(bp); continue; } if (lvp == NULL) { unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); } else { ASSERT_VOP_LOCKED(vp, "getbuf"); unlock = false; error = VOP_ISLOCKED(vp) == LK_EXCLUSIVE ? 0 : vn_lock(vp, LK_TRYUPGRADE); } if (error == 0) { CTR3(KTR_BUF, "flushbufqueue(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (curproc == bufdaemonproc) { vfs_bio_awrite(bp); } else { bremfree(bp); bwrite(bp); notbufdflushes++; } vn_finished_write(mp); if (unlock) VOP_UNLOCK(vp, 0); flushwithdeps += hasdeps; flushed++; /* * Sleeping on runningbufspace while holding * vnode lock leads to deadlock. */ if (curproc == bufdaemonproc && runningbufspace > hirunningspace) waitrunningbufspace(); continue; } vn_finished_write(mp); BUF_UNLOCK(bp); } mtx_lock(&bqlocks[queue]); TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); free(sentinel, M_TEMP); return (flushed); } /* * Check to see if a block is currently memory resident. */ struct buf * incore(struct bufobj *bo, daddr_t blkno) { struct buf *bp; BO_RLOCK(bo); bp = gbincore(bo, blkno); BO_RUNLOCK(bo); return (bp); } /* * Returns true if no I/O is needed to access the * associated VM object. This is like incore except * it also hunts around in the VM system for the data. */ static int inmem(struct vnode * vp, daddr_t blkno) { vm_object_t obj; vm_offset_t toff, tinc, size; vm_page_t m; vm_ooffset_t off; ASSERT_VOP_LOCKED(vp, "inmem"); if (incore(&vp->v_bufobj, blkno)) return 1; if (vp->v_mount == NULL) return 0; obj = vp->v_object; if (obj == NULL) return (0); size = PAGE_SIZE; if (size > vp->v_mount->mnt_stat.f_iosize) size = vp->v_mount->mnt_stat.f_iosize; off = (vm_ooffset_t)blkno * (vm_ooffset_t)vp->v_mount->mnt_stat.f_iosize; VM_OBJECT_RLOCK(obj); for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) { m = vm_page_lookup(obj, OFF_TO_IDX(off + toff)); if (!m) goto notinmem; tinc = size; if (tinc > PAGE_SIZE - ((toff + off) & PAGE_MASK)) tinc = PAGE_SIZE - ((toff + off) & PAGE_MASK); if (vm_page_is_valid(m, (vm_offset_t) ((toff + off) & PAGE_MASK), tinc) == 0) goto notinmem; } VM_OBJECT_RUNLOCK(obj); return 1; notinmem: VM_OBJECT_RUNLOCK(obj); return (0); } /* * Set the dirty range for a buffer based on the status of the dirty * bits in the pages comprising the buffer. The range is limited * to the size of the buffer. * * Tell the VM system that the pages associated with this buffer * are clean. This is used for delayed writes where the data is * going to go to disk eventually without additional VM intevention. * * Note that while we only really need to clean through to b_bcount, we * just go ahead and clean through to b_bufsize. */ static void vfs_clean_pages_dirty_buf(struct buf *bp) { vm_ooffset_t foff, noff, eoff; vm_page_t m; int i; if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0) return; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_clean_pages_dirty_buf: no buffer offset")); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); vfs_drain_busy_pages(bp); vfs_setdirty_locked_object(bp); for (i = 0; i < bp->b_npages; i++) { noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; eoff = noff; if (eoff > bp->b_offset + bp->b_bufsize) eoff = bp->b_offset + bp->b_bufsize; m = bp->b_pages[i]; vfs_page_set_validclean(bp, foff, m); /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ foff = noff; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } static void vfs_setdirty_locked_object(struct buf *bp) { vm_object_t object; int i; object = bp->b_bufobj->bo_object; VM_OBJECT_ASSERT_WLOCKED(object); /* * We qualify the scan for modified pages on whether the * object has been flushed yet. */ if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) { vm_offset_t boffset; vm_offset_t eoffset; /* * test the pages to see if they have been modified directly * by users through the VM system. */ for (i = 0; i < bp->b_npages; i++) vm_page_test_dirty(bp->b_pages[i]); /* * Calculate the encompassing dirty range, boffset and eoffset, * (eoffset - boffset) bytes. */ for (i = 0; i < bp->b_npages; i++) { if (bp->b_pages[i]->dirty) break; } boffset = (i << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); for (i = bp->b_npages - 1; i >= 0; --i) { if (bp->b_pages[i]->dirty) { break; } } eoffset = ((i + 1) << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); /* * Fit it to the buffer. */ if (eoffset > bp->b_bcount) eoffset = bp->b_bcount; /* * If we have a good dirty range, merge with the existing * dirty range. */ if (boffset < eoffset) { if (bp->b_dirtyoff > boffset) bp->b_dirtyoff = boffset; if (bp->b_dirtyend < eoffset) bp->b_dirtyend = eoffset; } } } /* * Allocate the KVA mapping for an existing buffer. * If an unmapped buffer is provided but a mapped buffer is requested, take * also care to properly setup mappings between pages and KVA. */ static void bp_unmapped_get_kva(struct buf *bp, daddr_t blkno, int size, int gbflags) { int bsize, maxsize, need_mapping, need_kva; off_t offset; need_mapping = bp->b_data == unmapped_buf && (gbflags & GB_UNMAPPED) == 0; need_kva = bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf && (gbflags & GB_KVAALLOC) != 0; if (!need_mapping && !need_kva) return; BUF_CHECK_UNMAPPED(bp); if (need_mapping && bp->b_kvabase != unmapped_buf) { /* * Buffer is not mapped, but the KVA was already * reserved at the time of the instantiation. Use the * allocated space. */ goto has_addr; } /* * Calculate the amount of the address space we would reserve * if the buffer was mapped. */ bsize = vn_isdisk(bp->b_vp, NULL) ? DEV_BSIZE : bp->b_bufobj->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; maxsize = size + (offset & PAGE_MASK); maxsize = imax(maxsize, bsize); while (bufkva_alloc(bp, maxsize, gbflags) != 0) { if ((gbflags & GB_NOWAIT_BD) != 0) { /* * XXXKIB: defragmentation cannot * succeed, not sure what else to do. */ panic("GB_NOWAIT_BD and GB_UNMAPPED %p", bp); } atomic_add_int(&mappingrestarts, 1); bufspace_wait(bp->b_vp, gbflags, 0, 0); } has_addr: if (need_mapping) { /* b_offset is handled by bpmap_qenter. */ bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); bpmap_qenter(bp); } } /* * getblk: * * Get a block given a specified block and offset into a file/device. * The buffers B_DONE bit will be cleared on return, making it almost * ready for an I/O initiation. B_INVAL may or may not be set on * return. The caller should clear B_INVAL prior to initiating a * READ. * * For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for * an existing buffer. * * For a VMIO buffer, B_CACHE is modified according to the backing VM. * If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set * and then cleared based on the backing VM. If the previous buffer is * non-0-sized but invalid, B_CACHE will be cleared. * * If getblk() must create a new buffer, the new buffer is returned with * both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which * case it is returned with B_INVAL clear and B_CACHE set based on the * backing VM. * * getblk() also forces a bwrite() for any B_DELWRI buffer whos * B_CACHE bit is clear. * * What this means, basically, is that the caller should use B_CACHE to * determine whether the buffer is fully valid or not and should clear * B_INVAL prior to issuing a read. If the caller intends to validate * the buffer by loading its data area with something, the caller needs * to clear B_INVAL. If the caller does this without issuing an I/O, * the caller should set B_CACHE ( as an optimization ), else the caller * should issue the I/O and biodone() will set B_CACHE if the I/O was * a write attempt or if it was a successful read. If the caller * intends to issue a READ, the caller must clear B_INVAL and BIO_ERROR * prior to issuing the READ. biodone() will *not* clear B_INVAL. */ struct buf * getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo, int flags) { struct buf *bp; struct bufobj *bo; int bsize, error, maxsize, vmio; off_t offset; CTR3(KTR_BUF, "getblk(%p, %ld, %d)", vp, (long)blkno, size); KASSERT((flags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); ASSERT_VOP_LOCKED(vp, "getblk"); - if (size > MAXBCACHEBUF) - panic("getblk: size(%d) > MAXBCACHEBUF(%d)\n", size, - MAXBCACHEBUF); + if (size > maxbcachebuf) + panic("getblk: size(%d) > maxbcachebuf(%d)\n", size, + maxbcachebuf); if (!unmapped_buf_allowed) flags &= ~(GB_UNMAPPED | GB_KVAALLOC); bo = &vp->v_bufobj; loop: BO_RLOCK(bo); bp = gbincore(bo, blkno); if (bp != NULL) { int lockflags; /* * Buffer is in-core. If the buffer is not busy nor managed, * it must be on a queue. */ lockflags = LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK; if (flags & GB_LOCK_NOWAIT) lockflags |= LK_NOWAIT; error = BUF_TIMELOCK(bp, lockflags, BO_LOCKPTR(bo), "getblk", slpflag, slptimeo); /* * If we slept and got the lock we have to restart in case * the buffer changed identities. */ if (error == ENOLCK) goto loop; /* We timed out or were interrupted. */ else if (error) return (NULL); /* If recursed, assume caller knows the rules. */ else if (BUF_LOCKRECURSED(bp)) goto end; /* * The buffer is locked. B_CACHE is cleared if the buffer is * invalid. Otherwise, for a non-VMIO buffer, B_CACHE is set * and for a VMIO buffer B_CACHE is adjusted according to the * backing VM cache. */ if (bp->b_flags & B_INVAL) bp->b_flags &= ~B_CACHE; else if ((bp->b_flags & (B_VMIO | B_INVAL)) == 0) bp->b_flags |= B_CACHE; if (bp->b_flags & B_MANAGED) MPASS(bp->b_qindex == QUEUE_NONE); else bremfree(bp); /* * check for size inconsistencies for non-VMIO case. */ if (bp->b_bcount != size) { if ((bp->b_flags & B_VMIO) == 0 || (size > bp->b_kvasize)) { if (bp->b_flags & B_DELWRI) { bp->b_flags |= B_NOCACHE; bwrite(bp); } else { if (LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; brelse(bp); } else { bp->b_flags |= B_NOCACHE; bwrite(bp); } } goto loop; } } /* * Handle the case of unmapped buffer which should * become mapped, or the buffer for which KVA * reservation is requested. */ bp_unmapped_get_kva(bp, blkno, size, flags); /* * If the size is inconsistent in the VMIO case, we can resize * the buffer. This might lead to B_CACHE getting set or * cleared. If the size has not changed, B_CACHE remains * unchanged from its previous state. */ allocbuf(bp, size); KASSERT(bp->b_offset != NOOFFSET, ("getblk: no buffer offset")); /* * A buffer with B_DELWRI set and B_CACHE clear must * be committed before we can return the buffer in * order to prevent the caller from issuing a read * ( due to B_CACHE not being set ) and overwriting * it. * * Most callers, including NFS and FFS, need this to * operate properly either because they assume they * can issue a read if B_CACHE is not set, or because * ( for example ) an uncached B_DELWRI might loop due * to softupdates re-dirtying the buffer. In the latter * case, B_CACHE is set after the first write completes, * preventing further loops. * NOTE! b*write() sets B_CACHE. If we cleared B_CACHE * above while extending the buffer, we cannot allow the * buffer to remain with B_CACHE set after the write * completes or it will represent a corrupt state. To * deal with this we set B_NOCACHE to scrap the buffer * after the write. * * We might be able to do something fancy, like setting * B_CACHE in bwrite() except if B_DELWRI is already set, * so the below call doesn't set B_CACHE, but that gets real * confusing. This is much easier. */ if ((bp->b_flags & (B_CACHE|B_DELWRI)) == B_DELWRI) { bp->b_flags |= B_NOCACHE; bwrite(bp); goto loop; } bp->b_flags &= ~B_DONE; } else { /* * Buffer is not in-core, create new buffer. The buffer * returned by getnewbuf() is locked. Note that the returned * buffer is also considered valid (not marked B_INVAL). */ BO_RUNLOCK(bo); /* * If the user does not want us to create the buffer, bail out * here. */ if (flags & GB_NOCREAT) return NULL; if (numfreebuffers == 0 && TD_IS_IDLETHREAD(curthread)) return NULL; bsize = vn_isdisk(vp, NULL) ? DEV_BSIZE : bo->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; vmio = vp->v_object != NULL; if (vmio) { maxsize = size + (offset & PAGE_MASK); } else { maxsize = size; /* Do not allow non-VMIO notmapped buffers. */ flags &= ~(GB_UNMAPPED | GB_KVAALLOC); } maxsize = imax(maxsize, bsize); bp = getnewbuf(vp, slpflag, slptimeo, maxsize, flags); if (bp == NULL) { if (slpflag || slptimeo) return NULL; /* * XXX This is here until the sleep path is diagnosed * enough to work under very low memory conditions. * * There's an issue on low memory, 4BSD+non-preempt * systems (eg MIPS routers with 32MB RAM) where buffer * exhaustion occurs without sleeping for buffer * reclaimation. This just sticks in a loop and * constantly attempts to allocate a buffer, which * hits exhaustion and tries to wakeup bufdaemon. * This never happens because we never yield. * * The real solution is to identify and fix these cases * so we aren't effectively busy-waiting in a loop * until the reclaimation path has cycles to run. */ kern_yield(PRI_USER); goto loop; } /* * This code is used to make sure that a buffer is not * created while the getnewbuf routine is blocked. * This can be a problem whether the vnode is locked or not. * If the buffer is created out from under us, we have to * throw away the one we just created. * * Note: this must occur before we associate the buffer * with the vp especially considering limitations in * the splay tree implementation when dealing with duplicate * lblkno's. */ BO_LOCK(bo); if (gbincore(bo, blkno)) { BO_UNLOCK(bo); bp->b_flags |= B_INVAL; brelse(bp); bufspace_release(maxsize); goto loop; } /* * Insert the buffer into the hash, so that it can * be found by incore. */ bp->b_blkno = bp->b_lblkno = blkno; bp->b_offset = offset; bgetvp(vp, bp); BO_UNLOCK(bo); /* * set B_VMIO bit. allocbuf() the buffer bigger. Since the * buffer size starts out as 0, B_CACHE will be set by * allocbuf() for the VMIO case prior to it testing the * backing store for validity. */ if (vmio) { bp->b_flags |= B_VMIO; KASSERT(vp->v_object == bp->b_bufobj->bo_object, ("ARGH! different b_bufobj->bo_object %p %p %p\n", bp, vp->v_object, bp->b_bufobj->bo_object)); } else { bp->b_flags &= ~B_VMIO; KASSERT(bp->b_bufobj->bo_object == NULL, ("ARGH! has b_bufobj->bo_object %p %p\n", bp, bp->b_bufobj->bo_object)); BUF_CHECK_MAPPED(bp); } allocbuf(bp, size); bufspace_release(maxsize); bp->b_flags &= ~B_DONE; } CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp); BUF_ASSERT_HELD(bp); end: buf_track(bp, __func__); KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); return (bp); } /* * Get an empty, disassociated buffer of given size. The buffer is initially * set to B_INVAL. */ struct buf * geteblk(int size, int flags) { struct buf *bp; int maxsize; maxsize = (size + BKVAMASK) & ~BKVAMASK; while ((bp = getnewbuf(NULL, 0, 0, maxsize, flags)) == NULL) { if ((flags & GB_NOWAIT_BD) && (curthread->td_pflags & TDP_BUFNEED) != 0) return (NULL); } allocbuf(bp, size); bufspace_release(maxsize); bp->b_flags |= B_INVAL; /* b_dep cleared by getnewbuf() */ BUF_ASSERT_HELD(bp); return (bp); } /* * Truncate the backing store for a non-vmio buffer. */ static void vfs_nonvmio_truncate(struct buf *bp, int newbsize) { if (bp->b_flags & B_MALLOC) { /* * malloced buffers are not shrunk */ if (newbsize == 0) { bufmallocadjust(bp, 0); free(bp->b_data, M_BIOBUF); bp->b_data = bp->b_kvabase; bp->b_flags &= ~B_MALLOC; } return; } vm_hold_free_pages(bp, newbsize); bufspace_adjust(bp, newbsize); } /* * Extend the backing for a non-VMIO buffer. */ static void vfs_nonvmio_extend(struct buf *bp, int newbsize) { caddr_t origbuf; int origbufsize; /* * We only use malloced memory on the first allocation. * and revert to page-allocated memory when the buffer * grows. * * There is a potential smp race here that could lead * to bufmallocspace slightly passing the max. It * is probably extremely rare and not worth worrying * over. */ if (bp->b_bufsize == 0 && newbsize <= PAGE_SIZE/2 && bufmallocspace < maxbufmallocspace) { bp->b_data = malloc(newbsize, M_BIOBUF, M_WAITOK); bp->b_flags |= B_MALLOC; bufmallocadjust(bp, newbsize); return; } /* * If the buffer is growing on its other-than-first * allocation then we revert to the page-allocation * scheme. */ origbuf = NULL; origbufsize = 0; if (bp->b_flags & B_MALLOC) { origbuf = bp->b_data; origbufsize = bp->b_bufsize; bp->b_data = bp->b_kvabase; bufmallocadjust(bp, 0); bp->b_flags &= ~B_MALLOC; newbsize = round_page(newbsize); } vm_hold_load_pages(bp, (vm_offset_t) bp->b_data + bp->b_bufsize, (vm_offset_t) bp->b_data + newbsize); if (origbuf != NULL) { bcopy(origbuf, bp->b_data, origbufsize); free(origbuf, M_BIOBUF); } bufspace_adjust(bp, newbsize); } /* * This code constitutes the buffer memory from either anonymous system * memory (in the case of non-VMIO operations) or from an associated * VM object (in the case of VMIO operations). This code is able to * resize a buffer up or down. * * Note that this code is tricky, and has many complications to resolve * deadlock or inconsistent data situations. Tread lightly!!! * There are B_CACHE and B_DELWRI interactions that must be dealt with by * the caller. Calling this code willy nilly can result in the loss of data. * * allocbuf() only adjusts B_CACHE for VMIO buffers. getblk() deals with * B_CACHE for the non-VMIO case. */ int allocbuf(struct buf *bp, int size) { int newbsize; BUF_ASSERT_HELD(bp); if (bp->b_bcount == size) return (1); if (bp->b_kvasize != 0 && bp->b_kvasize < size) panic("allocbuf: buffer too small"); newbsize = roundup2(size, DEV_BSIZE); if ((bp->b_flags & B_VMIO) == 0) { if ((bp->b_flags & B_MALLOC) == 0) newbsize = round_page(newbsize); /* * Just get anonymous memory from the kernel. Don't * mess with B_CACHE. */ if (newbsize < bp->b_bufsize) vfs_nonvmio_truncate(bp, newbsize); else if (newbsize > bp->b_bufsize) vfs_nonvmio_extend(bp, newbsize); } else { int desiredpages; desiredpages = (size == 0) ? 0 : num_pages((bp->b_offset & PAGE_MASK) + newbsize); if (bp->b_flags & B_MALLOC) panic("allocbuf: VMIO buffer can't be malloced"); /* * Set B_CACHE initially if buffer is 0 length or will become * 0-length. */ if (size == 0 || bp->b_bufsize == 0) bp->b_flags |= B_CACHE; if (newbsize < bp->b_bufsize) vfs_vmio_truncate(bp, desiredpages); /* XXX This looks as if it should be newbsize > b_bufsize */ else if (size > bp->b_bcount) vfs_vmio_extend(bp, desiredpages, size); bufspace_adjust(bp, newbsize); } bp->b_bcount = size; /* requested buffer size. */ return (1); } extern int inflight_transient_maps; void biodone(struct bio *bp) { struct mtx *mtxp; void (*done)(struct bio *); vm_offset_t start, end; biotrack(bp, __func__); if ((bp->bio_flags & BIO_TRANSIENT_MAPPING) != 0) { bp->bio_flags &= ~BIO_TRANSIENT_MAPPING; bp->bio_flags |= BIO_UNMAPPED; start = trunc_page((vm_offset_t)bp->bio_data); end = round_page((vm_offset_t)bp->bio_data + bp->bio_length); bp->bio_data = unmapped_buf; pmap_qremove(start, atop(end - start)); vmem_free(transient_arena, start, end - start); atomic_add_int(&inflight_transient_maps, -1); } done = bp->bio_done; if (done == NULL) { mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->bio_flags |= BIO_DONE; wakeup(bp); mtx_unlock(mtxp); } else done(bp); } /* * Wait for a BIO to finish. */ int biowait(struct bio *bp, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->bio_flags & BIO_DONE) == 0) msleep(bp, mtxp, PRIBIO, wchan, 0); mtx_unlock(mtxp); if (bp->bio_error != 0) return (bp->bio_error); if (!(bp->bio_flags & BIO_ERROR)) return (0); return (EIO); } void biofinish(struct bio *bp, struct devstat *stat, int error) { if (error) { bp->bio_error = error; bp->bio_flags |= BIO_ERROR; } if (stat != NULL) devstat_end_transaction_bio(stat, bp); biodone(bp); } #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING) void biotrack_buf(struct bio *bp, const char *location) { buf_track(bp->bio_track_bp, location); } #endif /* * bufwait: * * Wait for buffer I/O completion, returning error status. The buffer * is left locked and B_DONE on return. B_EINTR is converted into an EINTR * error and cleared. */ int bufwait(struct buf *bp) { if (bp->b_iocmd == BIO_READ) bwait(bp, PRIBIO, "biord"); else bwait(bp, PRIBIO, "biowr"); if (bp->b_flags & B_EINTR) { bp->b_flags &= ~B_EINTR; return (EINTR); } if (bp->b_ioflags & BIO_ERROR) { return (bp->b_error ? bp->b_error : EIO); } else { return (0); } } /* * bufdone: * * Finish I/O on a buffer, optionally calling a completion function. * This is usually called from an interrupt so process blocking is * not allowed. * * biodone is also responsible for setting B_CACHE in a B_VMIO bp. * In a non-VMIO bp, B_CACHE will be set on the next getblk() * assuming B_INVAL is clear. * * For the VMIO case, we set B_CACHE if the op was a read and no * read error occurred, or if the op was a write. B_CACHE is never * set if the buffer is invalid or otherwise uncacheable. * * biodone does not mess with B_INVAL, allowing the I/O routine or the * initiator to leave B_INVAL set to brelse the buffer out of existence * in the biodone routine. */ void bufdone(struct buf *bp) { struct bufobj *dropobj; void (*biodone)(struct buf *); buf_track(bp, __func__); CTR3(KTR_BUF, "bufdone(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); dropobj = NULL; KASSERT(!(bp->b_flags & B_DONE), ("biodone: bp %p already done", bp)); BUF_ASSERT_HELD(bp); runningbufwakeup(bp); if (bp->b_iocmd == BIO_WRITE) dropobj = bp->b_bufobj; /* call optional completion function if requested */ if (bp->b_iodone != NULL) { biodone = bp->b_iodone; bp->b_iodone = NULL; (*biodone) (bp); if (dropobj) bufobj_wdrop(dropobj); return; } bufdone_finish(bp); if (dropobj) bufobj_wdrop(dropobj); } void bufdone_finish(struct buf *bp) { BUF_ASSERT_HELD(bp); if (!LIST_EMPTY(&bp->b_dep)) buf_complete(bp); if (bp->b_flags & B_VMIO) { /* * Set B_CACHE if the op was a normal read and no error * occurred. B_CACHE is set for writes in the b*write() * routines. */ if (bp->b_iocmd == BIO_READ && !(bp->b_flags & (B_INVAL|B_NOCACHE)) && !(bp->b_ioflags & BIO_ERROR)) bp->b_flags |= B_CACHE; vfs_vmio_iodone(bp); } /* * For asynchronous completions, release the buffer now. The brelse * will do a wakeup there if necessary - so no need to do a wakeup * here in the async case. The sync case always needs to do a wakeup. */ if (bp->b_flags & B_ASYNC) { if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_RELBUF)) || (bp->b_ioflags & BIO_ERROR)) brelse(bp); else bqrelse(bp); } else bdone(bp); } /* * This routine is called in lieu of iodone in the case of * incomplete I/O. This keeps the busy status for pages * consistent. */ void vfs_unbusy_pages(struct buf *bp) { int i; vm_object_t obj; vm_page_t m; runningbufwakeup(bp); if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) { m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_offset) + i); if (!m) panic("vfs_unbusy_pages: page missing\n"); bp->b_pages[i] = m; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); } vm_page_sunbusy(m); } vm_object_pip_wakeupn(obj, bp->b_npages); VM_OBJECT_WUNLOCK(obj); } /* * vfs_page_set_valid: * * Set the valid bits in a page based on the supplied offset. The * range is restricted to the buffer's size. * * This routine is typically called after a read completes. */ static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t eoff; /* * Compute the end offset, eoff, such that [off, eoff) does not span a * page boundary and eoff is not greater than the end of the buffer. * The end of the buffer, in this case, is our file EOF, not the * allocation size of the buffer. */ eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > off) vm_page_set_valid_range(m, off & PAGE_MASK, eoff - off); } /* * vfs_page_set_validclean: * * Set the valid bits and clear the dirty bits in a page based on the * supplied offset. The range is restricted to the buffer's size. */ static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t soff, eoff; /* * Start and end offsets in buffer. eoff - soff may not cross a * page boundary or cross the end of the buffer. The end of the * buffer, in this case, is our file EOF, not the allocation size * of the buffer. */ soff = off; eoff = (off + PAGE_SIZE) & ~(off_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > soff) { vm_page_set_validclean( m, (vm_offset_t) (soff & PAGE_MASK), (vm_offset_t) (eoff - soff) ); } } /* * Ensure that all buffer pages are not exclusive busied. If any page is * exclusive busy, drain it. */ void vfs_drain_busy_pages(struct buf *bp) { vm_page_t m; int i, last_busied; VM_OBJECT_ASSERT_WLOCKED(bp->b_bufobj->bo_object); last_busied = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (vm_page_xbusied(m)) { for (; last_busied < i; last_busied++) vm_page_sbusy(bp->b_pages[last_busied]); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); vm_page_busy_sleep(m, "vbpage", true); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); } } } for (i = 0; i < last_busied; i++) vm_page_sunbusy(bp->b_pages[i]); } /* * This routine is called before a device strategy routine. * It is used to tell the VM system that paging I/O is in * progress, and treat the pages associated with the buffer * almost as being exclusive busy. Also the object paging_in_progress * flag is handled to make sure that the object doesn't become * inconsistent. * * Since I/O has not been initiated yet, certain buffer flags * such as BIO_ERROR or B_INVAL may be in an inconsistent state * and should be ignored. */ void vfs_busy_pages(struct buf *bp, int clear_modify) { vm_object_t obj; vm_ooffset_t foff; vm_page_t m; int i; bool bogus; if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_busy_pages: no buffer offset")); VM_OBJECT_WLOCK(obj); vfs_drain_busy_pages(bp); if (bp->b_bufsize != 0) vfs_setdirty_locked_object(bp); bogus = false; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if ((bp->b_flags & B_CLUSTER) == 0) { vm_object_pip_add(obj, 1); vm_page_sbusy(m); } /* * When readying a buffer for a read ( i.e * clear_modify == 0 ), it is important to do * bogus_page replacement for valid pages in * partially instantiated buffers. Partially * instantiated buffers can, in turn, occur when * reconstituting a buffer from its VM backing store * base. We only have to do this if B_CACHE is * clear ( which causes the I/O to occur in the * first place ). The replacement prevents the read * I/O from overwriting potentially dirty VM-backed * pages. XXX bogus page replacement is, uh, bogus. * It may not work properly with small-block devices. * We need to find a better way. */ if (clear_modify) { pmap_remove_write(m); vfs_page_set_validclean(bp, foff, m); } else if (m->valid == VM_PAGE_BITS_ALL && (bp->b_flags & B_CACHE) == 0) { bp->b_pages[i] = bogus_page; bogus = true; } foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; } VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * vfs_bio_set_valid: * * Set the range within the buffer to valid. The range is * relative to the beginning of the buffer, b_offset. Note that * b_offset itself may be offset from the beginning of the first * page. */ void vfs_bio_set_valid(struct buf *bp, int base, int size) { int i, n; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; /* * Fixup base to be relative to beginning of first page. * Set initial n to be the maximum number of bytes in the * first page that can be validated. */ base += (bp->b_offset & PAGE_MASK); n = PAGE_SIZE - (base & PAGE_MASK); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; vm_page_set_valid_range(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } /* * vfs_bio_clrbuf: * * If the specified buffer is a non-VMIO buffer, clear the entire * buffer. If the specified buffer is a VMIO buffer, clear and * validate only the previously invalid portions of the buffer. * This routine essentially fakes an I/O, so we need to clear * BIO_ERROR and B_INVAL. * * Note that while we only theoretically need to clear through b_bcount, * we go ahead and clear through b_bufsize. */ void vfs_bio_clrbuf(struct buf *bp) { int i, j, mask, sa, ea, slide; if ((bp->b_flags & (B_VMIO | B_MALLOC)) != B_VMIO) { clrbuf(bp); return; } bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); if ((bp->b_npages == 1) && (bp->b_bufsize < PAGE_SIZE) && (bp->b_offset & PAGE_MASK) == 0) { if (bp->b_pages[0] == bogus_page) goto unlock; mask = (1 << (bp->b_bufsize / DEV_BSIZE)) - 1; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[0]->object); if ((bp->b_pages[0]->valid & mask) == mask) goto unlock; if ((bp->b_pages[0]->valid & mask) == 0) { pmap_zero_page_area(bp->b_pages[0], 0, bp->b_bufsize); bp->b_pages[0]->valid |= mask; goto unlock; } } sa = bp->b_offset & PAGE_MASK; slide = 0; for (i = 0; i < bp->b_npages; i++, sa = 0) { slide = imin(slide + PAGE_SIZE, bp->b_offset + bp->b_bufsize); ea = slide & PAGE_MASK; if (ea == 0) ea = PAGE_SIZE; if (bp->b_pages[i] == bogus_page) continue; j = sa / DEV_BSIZE; mask = ((1 << ((ea - sa) / DEV_BSIZE)) - 1) << j; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[i]->object); if ((bp->b_pages[i]->valid & mask) == mask) continue; if ((bp->b_pages[i]->valid & mask) == 0) pmap_zero_page_area(bp->b_pages[i], sa, ea - sa); else { for (; sa < ea; sa += DEV_BSIZE, j++) { if ((bp->b_pages[i]->valid & (1 << j)) == 0) { pmap_zero_page_area(bp->b_pages[i], sa, DEV_BSIZE); } } } bp->b_pages[i]->valid |= mask; } unlock: VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); bp->b_resid = 0; } void vfs_bio_bzero_buf(struct buf *bp, int base, int size) { vm_page_t m; int i, n; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); bzero(bp->b_data + base, size); } else { BUF_CHECK_UNMAPPED(bp); n = PAGE_SIZE - (base & PAGE_MASK); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; pmap_zero_page_area(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } } } /* * Update buffer flags based on I/O request parameters, optionally releasing the * buffer. If it's VMIO or direct I/O, the buffer pages are released to the VM, * where they may be placed on a page queue (VMIO) or freed immediately (direct * I/O). Otherwise the buffer is released to the cache. */ static void b_io_dismiss(struct buf *bp, int ioflag, bool release) { KASSERT((ioflag & IO_NOREUSE) == 0 || (ioflag & IO_VMIO) != 0, ("buf %p non-VMIO noreuse", bp)); if ((ioflag & IO_DIRECT) != 0) bp->b_flags |= B_DIRECT; if ((ioflag & (IO_VMIO | IO_DIRECT)) != 0 && LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; if ((ioflag & IO_NOREUSE) != 0) bp->b_flags |= B_NOREUSE; if (release) brelse(bp); } else if (release) bqrelse(bp); } void vfs_bio_brelse(struct buf *bp, int ioflag) { b_io_dismiss(bp, ioflag, true); } void vfs_bio_set_flags(struct buf *bp, int ioflag) { b_io_dismiss(bp, ioflag, false); } /* * vm_hold_load_pages and vm_hold_free_pages get pages into * a buffers address space. The pages are anonymous and are * not associated with a file object. */ static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to) { vm_offset_t pg; vm_page_t p; int index; BUF_CHECK_MAPPED(bp); to = round_page(to); from = round_page(from); index = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; for (pg = from; pg < to; pg += PAGE_SIZE, index++) { tryagain: /* * note: must allocate system pages since blocking here * could interfere with paging I/O, no matter which * process we are. */ p = vm_page_alloc(NULL, 0, VM_ALLOC_SYSTEM | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_COUNT((to - pg) >> PAGE_SHIFT)); if (p == NULL) { VM_WAIT; goto tryagain; } pmap_qenter(pg, &p, 1); bp->b_pages[index] = p; } bp->b_npages = index; } /* Return pages associated with this buf to the vm system */ static void vm_hold_free_pages(struct buf *bp, int newbsize) { vm_offset_t from; vm_page_t p; int index, newnpages; BUF_CHECK_MAPPED(bp); from = round_page((vm_offset_t)bp->b_data + newbsize); newnpages = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; if (bp->b_npages > newnpages) pmap_qremove(from, bp->b_npages - newnpages); for (index = newnpages; index < bp->b_npages; index++) { p = bp->b_pages[index]; bp->b_pages[index] = NULL; if (vm_page_sbusied(p)) printf("vm_hold_free_pages: blkno: %jd, lblkno: %jd\n", (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); p->wire_count--; vm_page_free(p); atomic_subtract_int(&vm_cnt.v_wire_count, 1); } bp->b_npages = newnpages; } /* * Map an IO request into kernel virtual address space. * * All requests are (re)mapped into kernel VA space. * Notice that we use b_bufsize for the size of the buffer * to be mapped. b_bcount might be modified by the driver. * * Note that even if the caller determines that the address space should * be valid, a race or a smaller-file mapped into a larger space may * actually cause vmapbuf() to fail, so all callers of vmapbuf() MUST * check the return value. * * This function only works with pager buffers. */ int vmapbuf(struct buf *bp, int mapbuf) { vm_prot_t prot; int pidx; if (bp->b_bufsize < 0) return (-1); prot = VM_PROT_READ; if (bp->b_iocmd == BIO_READ) prot |= VM_PROT_WRITE; /* Less backwards than it looks */ if ((pidx = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, (vm_offset_t)bp->b_data, bp->b_bufsize, prot, bp->b_pages, btoc(MAXPHYS))) < 0) return (-1); bp->b_npages = pidx; bp->b_offset = ((vm_offset_t)bp->b_data) & PAGE_MASK; if (mapbuf || !unmapped_buf_allowed) { pmap_qenter((vm_offset_t)bp->b_kvabase, bp->b_pages, pidx); bp->b_data = bp->b_kvabase + bp->b_offset; } else bp->b_data = unmapped_buf; return(0); } /* * Free the io map PTEs associated with this IO operation. * We also invalidate the TLB entries and restore the original b_addr. * * This function only works with pager buffers. */ void vunmapbuf(struct buf *bp) { int npages; npages = bp->b_npages; if (buf_mapped(bp)) pmap_qremove(trunc_page((vm_offset_t)bp->b_data), npages); vm_page_unhold_pages(bp->b_pages, npages); bp->b_data = unmapped_buf; } void bdone(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_flags |= B_DONE; wakeup(bp); mtx_unlock(mtxp); } void bwait(struct buf *bp, u_char pri, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->b_flags & B_DONE) == 0) msleep(bp, mtxp, pri, wchan, 0); mtx_unlock(mtxp); } int bufsync(struct bufobj *bo, int waitfor) { return (VOP_FSYNC(bo2vnode(bo), waitfor, curthread)); } void bufstrategy(struct bufobj *bo, struct buf *bp) { int i = 0; struct vnode *vp; vp = bp->b_vp; KASSERT(vp == bo->bo_private, ("Inconsistent vnode bufstrategy")); KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("Wrong vnode in bufstrategy(bp=%p, vp=%p)", bp, vp)); i = VOP_STRATEGY(vp, bp); KASSERT(i == 0, ("VOP_STRATEGY failed bp=%p vp=%p", bp, bp->b_vp)); } void bufobj_wrefl(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); ASSERT_BO_WLOCKED(bo); bo->bo_numoutput++; } void bufobj_wref(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); BO_LOCK(bo); bo->bo_numoutput++; BO_UNLOCK(bo); } void bufobj_wdrop(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wdrop")); BO_LOCK(bo); KASSERT(bo->bo_numoutput > 0, ("bufobj_wdrop non-positive count")); if ((--bo->bo_numoutput == 0) && (bo->bo_flag & BO_WWAIT)) { bo->bo_flag &= ~BO_WWAIT; wakeup(&bo->bo_numoutput); } BO_UNLOCK(bo); } int bufobj_wwait(struct bufobj *bo, int slpflag, int timeo) { int error; KASSERT(bo != NULL, ("NULL bo in bufobj_wwait")); ASSERT_BO_WLOCKED(bo); error = 0; while (bo->bo_numoutput) { bo->bo_flag |= BO_WWAIT; error = msleep(&bo->bo_numoutput, BO_LOCKPTR(bo), slpflag | (PRIBIO + 1), "bo_wwait", timeo); if (error) break; } return (error); } /* * Set bio_data or bio_ma for struct bio from the struct buf. */ void bdata2bio(struct buf *bp, struct bio *bip) { if (!buf_mapped(bp)) { KASSERT(unmapped_buf_allowed, ("unmapped")); bip->bio_ma = bp->b_pages; bip->bio_ma_n = bp->b_npages; bip->bio_data = unmapped_buf; bip->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bip->bio_flags |= BIO_UNMAPPED; KASSERT(round_page(bip->bio_ma_offset + bip->bio_length) / PAGE_SIZE == bp->b_npages, ("Buffer %p too short: %d %lld %d", bp, bip->bio_ma_offset, (long long)bip->bio_length, bip->bio_ma_n)); } else { bip->bio_data = bp->b_data; bip->bio_ma = NULL; } } /* * The MIPS pmap code currently doesn't handle aliased pages. * The VIPT caches may not handle page aliasing themselves, leading * to data corruption. * * As such, this code makes a system extremely unhappy if said * system doesn't support unaliasing the above situation in hardware. * Some "recent" systems (eg some mips24k/mips74k cores) don't enable * this feature at build time, so it has to be handled in software. * * Once the MIPS pmap/cache code grows to support this function on * earlier chips, it should be flipped back off. */ #ifdef __mips__ static int buf_pager_relbuf = 1; #else static int buf_pager_relbuf = 0; #endif SYSCTL_INT(_vfs, OID_AUTO, buf_pager_relbuf, CTLFLAG_RWTUN, &buf_pager_relbuf, 0, "Make buffer pager release buffers after reading"); /* * The buffer pager. It uses buffer reads to validate pages. * * In contrast to the generic local pager from vm/vnode_pager.c, this * pager correctly and easily handles volumes where the underlying * device block size is greater than the machine page size. The * buffer cache transparently extends the requested page run to be * aligned at the block boundary, and does the necessary bogus page * replacements in the addends to avoid obliterating already valid * pages. * * The only non-trivial issue is that the exclusive busy state for * pages, which is assumed by the vm_pager_getpages() interface, is * incompatible with the VMIO buffer cache's desire to share-busy the * pages. This function performs a trivial downgrade of the pages' * state before reading buffers, and a less trivial upgrade from the * shared-busy to excl-busy state after the read. */ int vfs_bio_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, int *rahead, vbg_get_lblkno_t get_lblkno, vbg_get_blksize_t get_blksize) { vm_page_t m; vm_object_t object; struct buf *bp; struct mount *mp; daddr_t lbn, lbnp; vm_ooffset_t la, lb, poff, poffe; long bsize; int bo_bs, br_flags, error, i, pgsin, pgsin_a, pgsin_b; bool redo, lpart; object = vp->v_object; mp = vp->v_mount; la = IDX_TO_OFF(ma[count - 1]->pindex); if (la >= object->un_pager.vnp.vnp_size) return (VM_PAGER_BAD); lpart = la + PAGE_SIZE > object->un_pager.vnp.vnp_size; bo_bs = get_blksize(vp, get_lblkno(vp, IDX_TO_OFF(ma[0]->pindex))); /* * Calculate read-ahead, behind and total pages. */ pgsin = count; lb = IDX_TO_OFF(ma[0]->pindex); pgsin_b = OFF_TO_IDX(lb - rounddown2(lb, bo_bs)); pgsin += pgsin_b; if (rbehind != NULL) *rbehind = pgsin_b; pgsin_a = OFF_TO_IDX(roundup2(la, bo_bs) - la); if (la + IDX_TO_OFF(pgsin_a) >= object->un_pager.vnp.vnp_size) pgsin_a = OFF_TO_IDX(roundup2(object->un_pager.vnp.vnp_size, PAGE_SIZE) - la); pgsin += pgsin_a; if (rahead != NULL) *rahead = pgsin_a; VM_CNT_INC(v_vnodein); VM_CNT_ADD(v_vnodepgsin, pgsin); br_flags = (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) ? GB_UNMAPPED : 0; VM_OBJECT_WLOCK(object); again: for (i = 0; i < count; i++) vm_page_busy_downgrade(ma[i]); VM_OBJECT_WUNLOCK(object); lbnp = -1; for (i = 0; i < count; i++) { m = ma[i]; /* * Pages are shared busy and the object lock is not * owned, which together allow for the pages' * invalidation. The racy test for validity avoids * useless creation of the buffer for the most typical * case when invalidation is not used in redo or for * parallel read. The shared->excl upgrade loop at * the end of the function catches the race in a * reliable way (protected by the object lock). */ if (m->valid == VM_PAGE_BITS_ALL) continue; poff = IDX_TO_OFF(m->pindex); poffe = MIN(poff + PAGE_SIZE, object->un_pager.vnp.vnp_size); for (; poff < poffe; poff += bsize) { lbn = get_lblkno(vp, poff); if (lbn == lbnp) goto next_page; lbnp = lbn; bsize = get_blksize(vp, lbn); error = bread_gb(vp, lbn, bsize, curthread->td_ucred, br_flags, &bp); if (error != 0) goto end_pages; if (LIST_EMPTY(&bp->b_dep)) { /* * Invalidation clears m->valid, but * may leave B_CACHE flag if the * buffer existed at the invalidation * time. In this case, recycle the * buffer to do real read on next * bread() after redo. * * Otherwise B_RELBUF is not strictly * necessary, enable to reduce buf * cache pressure. */ if (buf_pager_relbuf || m->valid != VM_PAGE_BITS_ALL) bp->b_flags |= B_RELBUF; bp->b_flags &= ~B_NOCACHE; brelse(bp); } else { bqrelse(bp); } } KASSERT(1 /* racy, enable for debugging */ || m->valid == VM_PAGE_BITS_ALL || i == count - 1, ("buf %d %p invalid", i, m)); if (i == count - 1 && lpart) { VM_OBJECT_WLOCK(object); if (m->valid != 0 && m->valid != VM_PAGE_BITS_ALL) vm_page_zero_invalid(m, TRUE); VM_OBJECT_WUNLOCK(object); } next_page:; } end_pages: VM_OBJECT_WLOCK(object); redo = false; for (i = 0; i < count; i++) { vm_page_sunbusy(ma[i]); ma[i] = vm_page_grab(object, ma[i]->pindex, VM_ALLOC_NORMAL); /* * Since the pages were only sbusy while neither the * buffer nor the object lock was held by us, or * reallocated while vm_page_grab() slept for busy * relinguish, they could have been invalidated. * Recheck the valid bits and re-read as needed. * * Note that the last page is made fully valid in the * read loop, and partial validity for the page at * index count - 1 could mean that the page was * invalidated or removed, so we must restart for * safety as well. */ if (ma[i]->valid != VM_PAGE_BITS_ALL) redo = true; } if (redo && error == 0) goto again; VM_OBJECT_WUNLOCK(object); return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); } #include "opt_ddb.h" #ifdef DDB #include /* DDB command to show buffer data */ DB_SHOW_COMMAND(buffer, db_show_buffer) { /* get args */ struct buf *bp = (struct buf *)addr; #ifdef FULL_BUF_TRACKING uint32_t i, j; #endif if (!have_addr) { db_printf("usage: show buffer \n"); return; } db_printf("buf at %p\n", bp); db_printf("b_flags = 0x%b, b_xflags=0x%b, b_vflags=0x%b\n", (u_int)bp->b_flags, PRINT_BUF_FLAGS, (u_int)bp->b_xflags, PRINT_BUF_XFLAGS, (u_int)bp->b_vflags, PRINT_BUF_VFLAGS); db_printf( "b_error = %d, b_bufsize = %ld, b_bcount = %ld, b_resid = %ld\n" "b_bufobj = (%p), b_data = %p, b_blkno = %jd, b_lblkno = %jd, " "b_dep = %p\n", bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid, bp->b_bufobj, bp->b_data, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno, bp->b_dep.lh_first); db_printf("b_kvabase = %p, b_kvasize = %d\n", bp->b_kvabase, bp->b_kvasize); if (bp->b_npages) { int i; db_printf("b_npages = %d, pages(OBJ, IDX, PA): ", bp->b_npages); for (i = 0; i < bp->b_npages; i++) { vm_page_t m; m = bp->b_pages[i]; if (m != NULL) db_printf("(%p, 0x%lx, 0x%lx)", m->object, (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m)); else db_printf("( ??? )"); if ((i + 1) < bp->b_npages) db_printf(","); } db_printf("\n"); } #if defined(FULL_BUF_TRACKING) db_printf("b_io_tracking: b_io_tcnt = %u\n", bp->b_io_tcnt); i = bp->b_io_tcnt % BUF_TRACKING_SIZE; for (j = 1; j <= BUF_TRACKING_SIZE; j++) { if (bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)] == NULL) continue; db_printf(" %2u: %s\n", j, bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)]); } #elif defined(BUF_TRACKING) db_printf("b_io_tracking: %s\n", bp->b_io_tracking); #endif db_printf(" "); BUF_LOCKPRINTINFO(bp); } DB_SHOW_COMMAND(lockedbufs, lockedbufs) { struct buf *bp; int i; for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (BUF_ISLOCKED(bp)) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); if (db_pager_quit) break; } } } DB_SHOW_COMMAND(vnodebufs, db_show_vnodebufs) { struct vnode *vp; struct buf *bp; if (!have_addr) { db_printf("usage: show vnodebufs \n"); return; } vp = (struct vnode *)addr; db_printf("Clean buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_clean.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } db_printf("Dirty buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } DB_COMMAND(countfreebufs, db_coundfreebufs) { struct buf *bp; int i, used = 0, nfree = 0; if (have_addr) { db_printf("usage: countfreebufs\n"); return; } for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (bp->b_qindex == QUEUE_EMPTY) nfree++; else used++; } db_printf("Counted %d free, %d used (%d tot)\n", nfree, used, nfree + used); db_printf("numfreebuffers is %d\n", numfreebuffers); } #endif /* DDB */ Index: head/sys/sys/param.h =================================================================== --- head/sys/sys/param.h (revision 320061) +++ head/sys/sys/param.h (revision 320062) @@ -1,363 +1,369 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * 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. * * @(#)param.h 8.3 (Berkeley) 4/4/95 * $FreeBSD$ */ #ifndef _SYS_PARAM_H_ #define _SYS_PARAM_H_ #include #define BSD 199506 /* System version (year & month). */ #define BSD4_3 1 #define BSD4_4 1 /* * __FreeBSD_version numbers are documented in the Porter's Handbook. * If you bump the version for any reason, you should update the documentation * there. * Currently this lives here in the doc/ repository: * * head/en_US.ISO8859-1/books/porters-handbook/versions/chapter.xml * * scheme is: Rxx * 'R' is in the range 0 to 4 if this is a release branch or * x.0-CURRENT before RELENG_*_0 is created, otherwise 'R' is * in the range 5 to 9. */ #undef __FreeBSD_version #define __FreeBSD_version 1200033 /* Master, propagated to newvers */ /* * __FreeBSD_kernel__ indicates that this system uses the kernel of FreeBSD, * which by definition is always true on FreeBSD. This macro is also defined * on other systems that use the kernel of FreeBSD, such as GNU/kFreeBSD. * * It is tempting to use this macro in userland code when we want to enable * kernel-specific routines, and in fact it's fine to do this in code that * is part of FreeBSD itself. However, be aware that as presence of this * macro is still not widespread (e.g. older FreeBSD versions, 3rd party * compilers, etc), it is STRONGLY DISCOURAGED to check for this macro in * external applications without also checking for __FreeBSD__ as an * alternative. */ #undef __FreeBSD_kernel__ #define __FreeBSD_kernel__ #ifdef _KERNEL #define P_OSREL_SIGWAIT 700000 #define P_OSREL_SIGSEGV 700004 #define P_OSREL_MAP_ANON 800104 #define P_OSREL_MAP_FSTRICT 1100036 #define P_OSREL_SHUTDOWN_ENOTCONN 1100077 #define P_OSREL_MAJOR(x) ((x) / 100000) #endif #ifndef LOCORE #include #endif /* * Machine-independent constants (some used in following include files). * Redefined constants are from POSIX 1003.1 limits file. * * MAXCOMLEN should be >= sizeof(ac_comm) (see ) */ #include #define MAXCOMLEN 19 /* max command name remembered */ #define MAXINTERP PATH_MAX /* max interpreter file name length */ #define MAXLOGNAME 33 /* max login name length (incl. NUL) */ #define MAXUPRC CHILD_MAX /* max simultaneous processes */ #define NCARGS ARG_MAX /* max bytes for an exec function */ #define NGROUPS (NGROUPS_MAX+1) /* max number groups */ #define NOFILE OPEN_MAX /* max open files per process */ #define NOGROUP 65535 /* marker for empty group set member */ #define MAXHOSTNAMELEN 256 /* max hostname size */ #define SPECNAMELEN 63 /* max length of devicename */ /* More types and definitions used throughout the kernel. */ #ifdef _KERNEL #include #include #ifndef LOCORE #include #include #endif #ifndef FALSE #define FALSE 0 #endif #ifndef TRUE #define TRUE 1 #endif #endif #ifndef _KERNEL /* Signals. */ #include #endif /* Machine type dependent parameters. */ #include #ifndef _KERNEL #include #endif #ifndef DEV_BSHIFT #define DEV_BSHIFT 9 /* log2(DEV_BSIZE) */ #endif #define DEV_BSIZE (1<>PAGE_SHIFT) #endif /* * btodb() is messy and perhaps slow because `bytes' may be an off_t. We * want to shift an unsigned type to avoid sign extension and we don't * want to widen `bytes' unnecessarily. Assume that the result fits in * a daddr_t. */ #ifndef btodb #define btodb(bytes) /* calculates (bytes / DEV_BSIZE) */ \ (sizeof (bytes) > sizeof(long) \ ? (daddr_t)((unsigned long long)(bytes) >> DEV_BSHIFT) \ : (daddr_t)((unsigned long)(bytes) >> DEV_BSHIFT)) #endif #ifndef dbtob #define dbtob(db) /* calculates (db * DEV_BSIZE) */ \ ((off_t)(db) << DEV_BSHIFT) #endif #define PRIMASK 0x0ff #define PCATCH 0x100 /* OR'd with pri for tsleep to check signals */ #define PDROP 0x200 /* OR'd with pri to stop re-entry of interlock mutex */ #define NZERO 0 /* default "nice" */ #define NBBY 8 /* number of bits in a byte */ #define NBPW sizeof(int) /* number of bytes per word (integer) */ #define CMASK 022 /* default file mask: S_IWGRP|S_IWOTH */ #define NODEV (dev_t)(-1) /* non-existent device */ /* * File system parameters and macros. * * MAXBSIZE - Filesystems are made out of blocks of at most MAXBSIZE bytes * per block. MAXBSIZE may be made larger without effecting * any existing filesystems as long as it does not exceed MAXPHYS, * and may be made smaller at the risk of not being able to use * filesystems which require a block size exceeding MAXBSIZE. * * MAXBCACHEBUF - Maximum size of a buffer in the buffer cache. This must * be >= MAXBSIZE and can be set differently for different * architectures by defining it in . * Making this larger allows NFS to do larger reads/writes. * * BKVASIZE - Nominal buffer space per buffer, in bytes. BKVASIZE is the * minimum KVM memory reservation the kernel is willing to make. * Filesystems can of course request smaller chunks. Actual * backing memory uses a chunk size of a page (PAGE_SIZE). * The default value here can be overridden on a per-architecture - * basis by defining it in . This should - * probably be done to increase its value, when MAXBCACHEBUF is - * defined as a larger value in . + * basis by defining it in . * * If you make BKVASIZE too small you risk seriously fragmenting * the buffer KVM map which may slow things down a bit. If you * make it too big the kernel will not be able to optimally use * the KVM memory reserved for the buffer cache and will wind * up with too-few buffers. * * The default is 16384, roughly 2x the block size used by a * normal UFS filesystem. */ #define MAXBSIZE 65536 /* must be power of 2 */ #ifndef MAXBCACHEBUF #define MAXBCACHEBUF MAXBSIZE /* must be a power of 2 >= MAXBSIZE */ #endif #ifndef BKVASIZE #define BKVASIZE 16384 /* must be power of 2 */ #endif #define BKVAMASK (BKVASIZE-1) + +/* + * This variable is tuned via vfs.maxbcachebuf and is set to the value of + * MAXBCACHEBUF by default. + */ +#ifdef _KERNEL +extern int maxbcachebuf; +#endif /* * MAXPATHLEN defines the longest permissible path length after expanding * symbolic links. It is used to allocate a temporary buffer from the buffer * pool in which to do the name expansion, hence should be a power of two, * and must be less than or equal to MAXBSIZE. MAXSYMLINKS defines the * maximum number of symbolic links that may be expanded in a path name. * It should be set high enough to allow all legitimate uses, but halt * infinite loops reasonably quickly. */ #define MAXPATHLEN PATH_MAX #define MAXSYMLINKS 32 /* Bit map related macros. */ #define setbit(a,i) (((unsigned char *)(a))[(i)/NBBY] |= 1<<((i)%NBBY)) #define clrbit(a,i) (((unsigned char *)(a))[(i)/NBBY] &= ~(1<<((i)%NBBY))) #define isset(a,i) \ (((const unsigned char *)(a))[(i)/NBBY] & (1<<((i)%NBBY))) #define isclr(a,i) \ ((((const unsigned char *)(a))[(i)/NBBY] & (1<<((i)%NBBY))) == 0) /* Macros for counting and rounding. */ #ifndef howmany #define howmany(x, y) (((x)+((y)-1))/(y)) #endif #define nitems(x) (sizeof((x)) / sizeof((x)[0])) #define rounddown(x, y) (((x)/(y))*(y)) #define rounddown2(x, y) ((x)&(~((y)-1))) /* if y is power of two */ #define roundup(x, y) ((((x)+((y)-1))/(y))*(y)) /* to any y */ #define roundup2(x, y) (((x)+((y)-1))&(~((y)-1))) /* if y is powers of two */ #define powerof2(x) ((((x)-1)&(x))==0) /* Macros for min/max. */ #define MIN(a,b) (((a)<(b))?(a):(b)) #define MAX(a,b) (((a)>(b))?(a):(b)) #ifdef _KERNEL /* * Basic byte order function prototypes for non-inline functions. */ #ifndef LOCORE #ifndef _BYTEORDER_PROTOTYPED #define _BYTEORDER_PROTOTYPED __BEGIN_DECLS __uint32_t htonl(__uint32_t); __uint16_t htons(__uint16_t); __uint32_t ntohl(__uint32_t); __uint16_t ntohs(__uint16_t); __END_DECLS #endif #endif #ifndef lint #ifndef _BYTEORDER_FUNC_DEFINED #define _BYTEORDER_FUNC_DEFINED #define htonl(x) __htonl(x) #define htons(x) __htons(x) #define ntohl(x) __ntohl(x) #define ntohs(x) __ntohs(x) #endif /* !_BYTEORDER_FUNC_DEFINED */ #endif /* lint */ #endif /* _KERNEL */ /* * Scale factor for scaled integers used to count %cpu time and load avgs. * * The number of CPU `tick's that map to a unique `%age' can be expressed * by the formula (1 / (2 ^ (FSHIFT - 11))). The maximum load average that * can be calculated (assuming 32 bits) can be closely approximated using * the formula (2 ^ (2 * (16 - FSHIFT))) for (FSHIFT < 15). * * For the scheduler to maintain a 1:1 mapping of CPU `tick' to `%age', * FSHIFT must be at least 11; this gives us a maximum load avg of ~1024. */ #define FSHIFT 11 /* bits to right of fixed binary point */ #define FSCALE (1<> (PAGE_SHIFT - DEV_BSHIFT)) #define ctodb(db) /* calculates pages to devblks */ \ ((db) << (PAGE_SHIFT - DEV_BSHIFT)) /* * Old spelling of __containerof(). */ #define member2struct(s, m, x) \ ((struct s *)(void *)((char *)(x) - offsetof(struct s, m))) /* * Access a variable length array that has been declared as a fixed * length array. */ #define __PAST_END(array, offset) (((__typeof__(*(array)) *)(array))[offset]) #endif /* _SYS_PARAM_H_ */