/*
- * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
extern int nfsrv_wg_delay_v3;
static int nfsrv_require_resv_port = 0;
-static int nfsrv_deadsock_timer_on = 0;
+static time_t nfsrv_idlesock_timer_on = 0;
+static int nfsrv_sock_tcp_cnt = 0;
+#define NFSD_MIN_IDLE_TIMEOUT 30
+static int nfsrv_sock_idle_timeout = 3600; /* One hour */
int nfssvc_export(user_addr_t argp);
int nfssvc_nfsd(void);
SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, access_for_getattr, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_access_for_getattr, 0, "");
SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, idmap_ctrl, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_idmap_ctrl, 0, "");
SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, callback_port, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_callback_port, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, is_mobile, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_is_mobile, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, squishy_flags, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_squishy_flags, 0, "");
+SYSCTL_UINT(_vfs_generic_nfs_client, OID_AUTO, debug_ctl, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_debug_ctl, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, readlink_nocache, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_readlink_nocache, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_client, OID_AUTO, root_steals_gss_context, CTLFLAG_RW | CTLFLAG_LOCKED, &nfs_root_steals_ctx, 0, "");
#endif /* NFSCLIENT */
#if NFSSERVER
#endif
SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, nfsd_thread_max, CTLFLAG_RW | CTLFLAG_LOCKED, &nfsd_thread_max, 0, "");
SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, nfsd_thread_count, CTLFLAG_RD | CTLFLAG_LOCKED, &nfsd_thread_count, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, nfsd_sock_idle_timeout, CTLFLAG_RW | CTLFLAG_LOCKED, &nfsrv_sock_idle_timeout, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, nfsd_tcp_connections, CTLFLAG_RD | CTLFLAG_LOCKED, &nfsrv_sock_tcp_cnt, 0, "");
+#ifdef NFS_UC_Q_DEBUG
+SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, use_upcall_svc, CTLFLAG_RW | CTLFLAG_LOCKED, &nfsrv_uc_use_proxy, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, upcall_queue_limit, CTLFLAG_RW | CTLFLAG_LOCKED, &nfsrv_uc_queue_limit, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, upcall_queue_max_seen, CTLFLAG_RW | CTLFLAG_LOCKED, &nfsrv_uc_queue_max_seen, 0, "");
+SYSCTL_INT(_vfs_generic_nfs_server, OID_AUTO, upcall_queue_count, CTLFLAG_RD | CTLFLAG_LOCKED, __DECONST(int *, &nfsrv_uc_queue_count), 0, "");
+#endif
#endif /* NFSSERVER */
#if NFSCLIENT
+static int
+mapname2id(struct nfs_testmapid *map)
+{
+ int error;
+
+ error = nfs4_id2guid(map->ntm_name, &map->ntm_guid, map->ntm_grpflag);
+ if (error)
+ return (error);
+
+ if (map->ntm_grpflag)
+ error = kauth_cred_guid2gid(&map->ntm_guid, (gid_t *)&map->ntm_id);
+ else
+ error = kauth_cred_guid2uid(&map->ntm_guid, (uid_t *)&map->ntm_id);
+
+ return (error);
+}
+
+static int
+mapid2name(struct nfs_testmapid *map)
+{
+ int error;
+ int len = sizeof(map->ntm_name);
+
+ if (map->ntm_grpflag)
+ error = kauth_cred_gid2guid((gid_t)map->ntm_id, &map->ntm_guid);
+ else
+ error = kauth_cred_uid2guid((uid_t)map->ntm_id, &map->ntm_guid);
+
+ if (error)
+ return (error);
+
+ error = nfs4_guid2id(&map->ntm_guid, map->ntm_name, &len, map->ntm_grpflag);
+
+ return (error);
+
+}
+
+
+static int
+nfsclnt_testidmap(proc_t p, user_addr_t argp)
+{
+ struct nfs_testmapid mapid;
+ int error, coerror;
+
+ /* Let root make this call. */
+ error = proc_suser(p);
+ if (error)
+ return (error);
+
+ error = copyin(argp, &mapid, sizeof(mapid));
+ if (error)
+ return (error);
+ if (mapid.ntm_name2id)
+ error = mapname2id(&mapid);
+ else
+ error = mapid2name(&mapid);
+
+ coerror = copyout(&mapid, argp, sizeof(mapid));
+
+ return (error ? error : coerror);
+}
+
int
nfsclnt(proc_t p, struct nfsclnt_args *uap, __unused int *retval)
{
case NFSCLNT_LOCKDNOTIFY:
error = nfslockdnotify(p, uap->argp);
break;
+ case NFSCLNT_TESTIDMAP:
+ error = nfsclnt_testidmap(p, uap->argp);
+ break;
default:
error = EINVAL;
}
return (error);
}
+
/*
* Asynchronous I/O threads for client NFS.
* They do read-ahead and write-behind operations on the block I/O cache.
worktodo:
while ((nmp = niod->niod_nmp)) {
+ if (nmp == NULL){
+ niod->niod_nmp = NULL;
+ break;
+ }
+
/*
* Service this mount's async I/O queue.
*
/* grab the current contents of the queue */
TAILQ_INIT(&iodq);
TAILQ_CONCAT(&iodq, &nmp->nm_iodq, r_achain);
+ /* Mark each iod request as being managed by an iod */
+ TAILQ_FOREACH(req, &iodq, r_achain) {
+ lck_mtx_lock(&req->r_mtx);
+ assert(!(req->r_flags & R_IOD));
+ req->r_flags |= R_IOD;
+ lck_mtx_unlock(&req->r_mtx);
+ }
lck_mtx_unlock(nfsiod_mutex);
/* process the queue */
lck_mtx_lock(nfsiod_mutex);
morework = !TAILQ_EMPTY(&nmp->nm_iodq);
if (!morework || !TAILQ_EMPTY(&nfsiodmounts)) {
- /* we're going to stop working on this mount */
- if (morework) /* mount still needs more work so queue it up */
+ /*
+ * we're going to stop working on this mount but if the
+ * mount still needs more work so queue it up
+ */
+ if (morework && nmp->nm_iodlink.tqe_next == NFSNOLIST)
TAILQ_INSERT_TAIL(&nfsiodmounts, nmp, nm_iodlink);
nmp->nm_niod = NULL;
niod->niod_nmp = NULL;
if (!niod->niod_nmp && !TAILQ_EMPTY(&nfsiodmounts)) {
niod->niod_nmp = TAILQ_FIRST(&nfsiodmounts);
TAILQ_REMOVE(&nfsiodmounts, niod->niod_nmp, nm_iodlink);
+ niod->niod_nmp->nm_iodlink.tqe_next = NFSNOLIST;
}
if (niod->niod_nmp)
goto worktodo;
return (error);
}
-extern struct fileops vnops;
+extern const struct fileops vnops;
/*
* syscall for the rpc.lockd to use to translate a NFS file handle into
fp = nfp;
fp->f_fglob->fg_flag = fmode & FMASK;
- fp->f_fglob->fg_type = DTYPE_VNODE;
fp->f_fglob->fg_ops = &vnops;
fp->f_fglob->fg_data = (caddr_t)vp;
type = F_FLOCK;
if ((fmode & FNONBLOCK) == 0)
type |= F_WAIT;
- if ((error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, type, ctx))) {
+ if ((error = VNOP_ADVLOCK(vp, (caddr_t)fp->f_fglob, F_SETLK, &lf, type, ctx, NULL))) {
struct vfs_context context = *vfs_context_current();
/* Modify local copy (to not damage thread copy) */
context.vc_ucred = fp->f_fglob->fg_cred;
{
struct nfsrv_sock *slp;
int error = 0, sodomain, sotype, soprotocol, on = 1;
+ int first;
struct timeval timeo;
/* make sure mbuf constants are set up */
}
/* Set protocol options and reserve some space (for UDP). */
- if (sotype == SOCK_STREAM)
+ if (sotype == SOCK_STREAM) {
+ error = nfsrv_check_exports_allow_address(mynam);
+ if (error)
+ return (error);
sock_setsockopt(so, SOL_SOCKET, SO_KEEPALIVE, &on, sizeof(on));
+ }
if ((sodomain == AF_INET) && (soprotocol == IPPROTO_TCP))
sock_setsockopt(so, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on));
if (sotype == SOCK_DGRAM) { /* set socket buffer sizes for UDP */
}
/* add the socket to the list */
+ first = TAILQ_EMPTY(&nfsrv_socklist);
TAILQ_INSERT_TAIL(&nfsrv_socklist, slp, ns_chain);
+ if (soprotocol == IPPROTO_TCP) {
+ nfsrv_sock_tcp_cnt++;
+ if (nfsrv_sock_idle_timeout < 0)
+ nfsrv_sock_idle_timeout = 0;
+ if (nfsrv_sock_idle_timeout && (nfsrv_sock_idle_timeout < NFSD_MIN_IDLE_TIMEOUT))
+ nfsrv_sock_idle_timeout = NFSD_MIN_IDLE_TIMEOUT;
+ /*
+ * Possibly start or stop the idle timer. We only start the idle timer when
+ * we have more than 2 * nfsd_thread_max connections. If the idle timer is
+ * on then we may need to turn it off based on the nvsrv_sock_idle_timeout or
+ * the number of connections.
+ */
+ if ((nfsrv_sock_tcp_cnt > 2 * nfsd_thread_max) || nfsrv_idlesock_timer_on) {
+ if (nfsrv_sock_idle_timeout == 0 || nfsrv_sock_tcp_cnt <= 2 * nfsd_thread_max) {
+ if (nfsrv_idlesock_timer_on) {
+ thread_call_cancel(nfsrv_idlesock_timer_call);
+ nfsrv_idlesock_timer_on = 0;
+ }
+ } else {
+ struct nfsrv_sock *old_slp;
+ struct timeval now;
+ time_t time_to_wait = nfsrv_sock_idle_timeout;
+ /*
+ * Get the oldest tcp socket and calculate the
+ * earliest time for the next idle timer to fire
+ * based on the possibly updated nfsrv_sock_idle_timeout
+ */
+ TAILQ_FOREACH(old_slp, &nfsrv_socklist, ns_chain) {
+ if (old_slp->ns_sotype == SOCK_STREAM) {
+ microuptime(&now);
+ time_to_wait -= now.tv_sec - old_slp->ns_timestamp;
+ if (time_to_wait < 1)
+ time_to_wait = 1;
+ break;
+ }
+ }
+ /*
+ * If we have a timer scheduled, but if its going to fire too late,
+ * turn it off.
+ */
+ if (nfsrv_idlesock_timer_on > now.tv_sec + time_to_wait) {
+ thread_call_cancel(nfsrv_idlesock_timer_call);
+ nfsrv_idlesock_timer_on = 0;
+ }
+ /* Schedule the idle thread if it isn't already */
+ if (!nfsrv_idlesock_timer_on) {
+ nfs_interval_timer_start(nfsrv_idlesock_timer_call, time_to_wait * 1000);
+ nfsrv_idlesock_timer_on = now.tv_sec + time_to_wait;
+ }
+ }
+ }
+ }
sock_retain(so); /* grab a retain count on the socket */
slp->ns_so = so;
slp->ns_sotype = sotype;
slp->ns_nam = mynam;
- /* set up the socket upcall */
- sock_setupcall(so, nfsrv_rcv, slp);
- /* just playin' it safe */
- sock_setsockopt(so, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on));
+ /* set up the socket up-call */
+ nfsrv_uc_addsock(slp, first);
/* mark that the socket is not in the nfsrv_sockwg list */
slp->ns_wgq.tqe_next = SLPNOLIST;
-
+
slp->ns_flag = SLP_VALID | SLP_NEEDQ;
nfsrv_wakenfsd(slp);
u_quad_t cur_usec;
struct timeval now;
struct vfs_context context;
+ struct timespec to;
#ifndef nolint
cacherep = RC_DOIT;
lck_mtx_lock(nfsd_mutex);
if (nfsd_thread_count++ == 0)
nfsrv_initcache(); /* Init the server request cache */
+
TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
lck_mtx_unlock(nfsd_mutex);
context.vc_thread = current_thread();
+ /* Set time out so that nfsd threads can wake up a see if they are still needed. */
+ to.tv_sec = 5;
+ to.tv_nsec = 0;
+
/*
* Loop getting rpc requests until SIGKILL.
*/
}
nfsd->nfsd_flag |= NFSD_WAITING;
TAILQ_INSERT_HEAD(&nfsd_queue, nfsd, nfsd_queue);
- error = msleep(nfsd, nfsd_mutex, PSOCK | PCATCH, "nfsd", NULL);
+ error = msleep(nfsd, nfsd_mutex, PSOCK | PCATCH, "nfsd", &to);
if (error) {
if (nfsd->nfsd_flag & NFSD_WAITING) {
TAILQ_REMOVE(&nfsd_queue, nfsd, nfsd_queue);
nfsd->nfsd_flag &= ~NFSD_WAITING;
}
+ if (error == EWOULDBLOCK)
+ continue;
goto done;
}
}
if (!nfsd->nfsd_slp && slp) {
/* we found a socket to work on, grab a reference */
slp->ns_sref++;
+ microuptime(&now);
+ slp->ns_timestamp = now.tv_sec;
+ /* We keep the socket list in least recently used order for reaping idle sockets */
+ TAILQ_REMOVE(&nfsrv_socklist, slp, ns_chain);
+ TAILQ_INSERT_TAIL(&nfsrv_socklist, slp, ns_chain);
nfsd->nfsd_slp = slp;
opcnt = 0;
/* and put it at the back of the work queue */
}
if (error) {
- OSAddAtomic(1, &nfsstats.srv_errs);
+ OSAddAtomic64(1, &nfsstats.srv_errs);
nfsrv_updatecache(nd, FALSE, mrep);
if (nd->nd_nam2) {
mbuf_freem(nd->nd_nam2);
}
break;
}
- OSAddAtomic(1, &nfsstats.srvrpccnt[nd->nd_procnum]);
+ OSAddAtomic64(1, &nfsstats.srvrpccnt[nd->nd_procnum]);
nfsrv_updatecache(nd, TRUE, mrep);
/* FALLTHRU */
if (so == NULL)
return;
+ sock_setupcall(so, NULL, NULL);
+ sock_shutdown(so, SHUT_RDWR);
+
/*
- * Attempt to deter future upcalls, but leave the
- * upcall info in place to avoid a race with the
- * networking code.
+ * Remove from the up-call queue
*/
- socket_lock(so, 1);
- so->so_rcv.sb_flags &= ~SB_UPCALL;
- socket_unlock(so, 1);
-
- sock_shutdown(so, SHUT_RDWR);
+ nfsrv_uc_dequeue(slp);
}
/*
* Derefence a server socket structure. If it has no more references and
* is no longer valid, you can throw it away.
*/
-void
-nfsrv_slpderef(struct nfsrv_sock *slp)
+static void
+nfsrv_slpderef_locked(struct nfsrv_sock *slp)
{
- struct timeval now;
-
- lck_mtx_lock(nfsd_mutex);
lck_rw_lock_exclusive(&slp->ns_rwlock);
slp->ns_sref--;
slp->ns_flag &= ~SLP_QUEUED;
}
lck_rw_done(&slp->ns_rwlock);
- lck_mtx_unlock(nfsd_mutex);
return;
}
TAILQ_REMOVE(&nfsrv_sockwork, slp, ns_svcq);
slp->ns_flag &= ~SLP_QUEUED;
}
+ lck_rw_done(&slp->ns_rwlock);
- /*
- * Queue the socket up for deletion
- * and start the timer to delete it
- * after it has been in limbo for
- * a while.
- */
- microuptime(&now);
- slp->ns_timestamp = now.tv_sec;
TAILQ_REMOVE(&nfsrv_socklist, slp, ns_chain);
- TAILQ_INSERT_TAIL(&nfsrv_deadsocklist, slp, ns_chain);
- if (!nfsrv_deadsock_timer_on) {
- nfsrv_deadsock_timer_on = 1;
- nfs_interval_timer_start(nfsrv_deadsock_timer_call,
- NFSRV_DEADSOCKDELAY * 1000);
- }
+ if (slp->ns_sotype == SOCK_STREAM)
+ nfsrv_sock_tcp_cnt--;
- lck_rw_done(&slp->ns_rwlock);
/* now remove from the write gather socket list */
if (slp->ns_wgq.tqe_next != SLPNOLIST) {
TAILQ_REMOVE(&nfsrv_sockwg, slp, ns_wgq);
slp->ns_wgq.tqe_next = SLPNOLIST;
}
+ nfsrv_slpfree(slp);
+}
+
+void
+nfsrv_slpderef(struct nfsrv_sock *slp)
+{
+ lck_mtx_lock(nfsd_mutex);
+ nfsrv_slpderef_locked(slp);
lck_mtx_unlock(nfsd_mutex);
}
/*
- * Check periodically for dead sockets pending delete.
- * If a socket has been dead for more than NFSRV_DEADSOCKDELAY
- * seconds then we assume it's safe to free.
+ * Check periodically for idle sockest if needed and
+ * zap them.
*/
void
-nfsrv_deadsock_timer(__unused void *param0, __unused void *param1)
+nfsrv_idlesock_timer(__unused void *param0, __unused void *param1)
{
- struct nfsrv_sock *slp;
+ struct nfsrv_sock *slp, *tslp;
struct timeval now;
- time_t time_to_wait;
+ time_t time_to_wait = nfsrv_sock_idle_timeout;
microuptime(&now);
lck_mtx_lock(nfsd_mutex);
- while ((slp = TAILQ_FIRST(&nfsrv_deadsocklist))) {
- if ((slp->ns_timestamp + NFSRV_DEADSOCKDELAY) > now.tv_sec)
- break;
- TAILQ_REMOVE(&nfsrv_deadsocklist, slp, ns_chain);
- nfsrv_slpfree(slp);
- }
- if (TAILQ_EMPTY(&nfsrv_deadsocklist)) {
- nfsrv_deadsock_timer_on = 0;
+ /* Turn off the timer if we're suppose to and get out */
+ if (nfsrv_sock_idle_timeout < NFSD_MIN_IDLE_TIMEOUT)
+ nfsrv_sock_idle_timeout = 0;
+ if ((nfsrv_sock_tcp_cnt <= 2 * nfsd_thread_max) || (nfsrv_sock_idle_timeout == 0)) {
+ nfsrv_idlesock_timer_on = 0;
lck_mtx_unlock(nfsd_mutex);
return;
}
- time_to_wait = (slp->ns_timestamp + NFSRV_DEADSOCKDELAY) - now.tv_sec;
- if (time_to_wait < 1)
- time_to_wait = 1;
- lck_mtx_unlock(nfsd_mutex);
+ TAILQ_FOREACH_SAFE(slp, &nfsrv_socklist, ns_chain, tslp) {
+ lck_rw_lock_exclusive(&slp->ns_rwlock);
+ /* Skip udp and referenced sockets */
+ if (slp->ns_sotype == SOCK_DGRAM || slp->ns_sref) {
+ lck_rw_done(&slp->ns_rwlock);
+ continue;
+ }
+ /*
+ * If this is the first non-referenced socket that hasn't idle out,
+ * use its time stamp to calculate the earlist time in the future
+ * to start the next invocation of the timer. Since the nfsrv_socklist
+ * is sorted oldest access to newest. Once we find the first one,
+ * we're done and break out of the loop.
+ */
+ if (((slp->ns_timestamp + nfsrv_sock_idle_timeout) > now.tv_sec) ||
+ nfsrv_sock_tcp_cnt <= 2 * nfsd_thread_max) {
+ time_to_wait -= now.tv_sec - slp->ns_timestamp;
+ if (time_to_wait < 1)
+ time_to_wait = 1;
+ lck_rw_done(&slp->ns_rwlock);
+ break;
+ }
+ /*
+ * Bump the ref count. nfsrv_slpderef below will destroy
+ * the socket, since nfsrv_zapsock has closed it.
+ */
+ slp->ns_sref++;
+ nfsrv_zapsock(slp);
+ lck_rw_done(&slp->ns_rwlock);
+ nfsrv_slpderef_locked(slp);
+ }
- nfs_interval_timer_start(nfsrv_deadsock_timer_call,
- time_to_wait * 1000);
+ /* Start ourself back up */
+ nfs_interval_timer_start(nfsrv_idlesock_timer_call, time_to_wait * 1000);
+ /* Remember when the next timer will fire for nfssvc_addsock. */
+ nfsrv_idlesock_timer_on = now.tv_sec + time_to_wait;
+ lck_mtx_unlock(nfsd_mutex);
}
/*
microuptime(&now);
for (slp = TAILQ_FIRST(&nfsrv_socklist); slp != 0; slp = nslp) {
nslp = TAILQ_NEXT(slp, ns_chain);
- if (slp->ns_flag & SLP_VALID) {
- lck_rw_lock_exclusive(&slp->ns_rwlock);
+ lck_rw_lock_exclusive(&slp->ns_rwlock);
+ slp->ns_sref++;
+ if (slp->ns_flag & SLP_VALID)
nfsrv_zapsock(slp);
- lck_rw_done(&slp->ns_rwlock);
- }
- if (slp->ns_flag & SLP_QUEUED) {
- if (slp->ns_flag & SLP_WAITQ)
- TAILQ_REMOVE(&nfsrv_sockwait, slp, ns_svcq);
- else
- TAILQ_REMOVE(&nfsrv_sockwork, slp, ns_svcq);
- slp->ns_flag &= ~SLP_QUEUED;
- }
- if (slp->ns_wgq.tqe_next != SLPNOLIST) {
- TAILQ_REMOVE(&nfsrv_sockwg, slp, ns_wgq);
- slp->ns_wgq.tqe_next = SLPNOLIST;
- }
- /* queue the socket up for deletion */
- slp->ns_timestamp = now.tv_sec;
- TAILQ_REMOVE(&nfsrv_socklist, slp, ns_chain);
- TAILQ_INSERT_TAIL(&nfsrv_deadsocklist, slp, ns_chain);
- if (!nfsrv_deadsock_timer_on) {
- nfsrv_deadsock_timer_on = 1;
- nfs_interval_timer_start(nfsrv_deadsock_timer_call,
- NFSRV_DEADSOCKDELAY * 1000);
- }
+ lck_rw_done(&slp->ns_rwlock);
+ nfsrv_slpderef_locked(slp);
}
-
+#
#if CONFIG_FSE
/*
* Flush pending file write fsevents
lck_mtx_unlock(nfsrv_fmod_mutex);
#endif
+ nfsrv_uc_cleanup(); /* Stop nfs socket up-call threads */
+
nfs_gss_svc_cleanup(); /* Remove any RPCSEC_GSS contexts */
nfsrv_cleancache(); /* And clear out server cache */
projects / apple / xnu.git / blobdiff