/*
* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
"Reserved");
int
-lnet_fail_nid (lnet_nid_t nid, unsigned int threshold)
+lnet_fail_nid(lnet_nid_t nid, unsigned int threshold)
{
- lnet_test_peer_t *tp;
- cfs_list_t *el;
- cfs_list_t *next;
- cfs_list_t cull;
-
- LASSERT (the_lnet.ln_init);
+ lnet_test_peer_t *tp;
+ struct list_head *el;
+ struct list_head *next;
+ struct list_head cull;
+ /* NB: use lnet_net_lock(0) to serialize operations on test peers */
if (threshold != 0) {
/* Adding a new entry */
LIBCFS_ALLOC(tp, sizeof(*tp));
tp->tp_nid = nid;
tp->tp_threshold = threshold;
- LNET_LOCK();
- cfs_list_add_tail (&tp->tp_list, &the_lnet.ln_test_peers);
- LNET_UNLOCK();
- return 0;
- }
+ lnet_net_lock(0);
+ list_add_tail(&tp->tp_list, &the_lnet.ln_test_peers);
+ lnet_net_unlock(0);
+ return 0;
+ }
- /* removing entries */
- CFS_INIT_LIST_HEAD (&cull);
+ /* removing entries */
+ INIT_LIST_HEAD(&cull);
- LNET_LOCK();
+ lnet_net_lock(0);
- cfs_list_for_each_safe (el, next, &the_lnet.ln_test_peers) {
- tp = cfs_list_entry (el, lnet_test_peer_t, tp_list);
+ list_for_each_safe(el, next, &the_lnet.ln_test_peers) {
+ tp = list_entry(el, lnet_test_peer_t, tp_list);
- if (tp->tp_threshold == 0 || /* needs culling anyway */
- nid == LNET_NID_ANY || /* removing all entries */
- tp->tp_nid == nid) /* matched this one */
- {
- cfs_list_del (&tp->tp_list);
- cfs_list_add (&tp->tp_list, &cull);
- }
- }
+ if (tp->tp_threshold == 0 || /* needs culling anyway */
+ nid == LNET_NID_ANY || /* removing all entries */
+ tp->tp_nid == nid) { /* matched this one */
+ list_del(&tp->tp_list);
+ list_add(&tp->tp_list, &cull);
+ }
+ }
- LNET_UNLOCK();
+ lnet_net_unlock(0);
- while (!cfs_list_empty (&cull)) {
- tp = cfs_list_entry (cull.next, lnet_test_peer_t, tp_list);
+ while (!list_empty(&cull)) {
+ tp = list_entry(cull.next, lnet_test_peer_t, tp_list);
- cfs_list_del (&tp->tp_list);
- LIBCFS_FREE(tp, sizeof (*tp));
- }
- return 0;
+ list_del(&tp->tp_list);
+ LIBCFS_FREE(tp, sizeof(*tp));
+ }
+ return 0;
}
static int
fail_peer (lnet_nid_t nid, int outgoing)
{
- lnet_test_peer_t *tp;
- cfs_list_t *el;
- cfs_list_t *next;
- cfs_list_t cull;
- int fail = 0;
-
- CFS_INIT_LIST_HEAD (&cull);
-
- LNET_LOCK();
-
- cfs_list_for_each_safe (el, next, &the_lnet.ln_test_peers) {
- tp = cfs_list_entry (el, lnet_test_peer_t, tp_list);
-
- if (tp->tp_threshold == 0) {
- /* zombie entry */
- if (outgoing) {
- /* only cull zombies on outgoing tests,
- * since we may be at interrupt priority on
- * incoming messages. */
- cfs_list_del (&tp->tp_list);
- cfs_list_add (&tp->tp_list, &cull);
- }
- continue;
- }
+ lnet_test_peer_t *tp;
+ struct list_head *el;
+ struct list_head *next;
+ struct list_head cull;
+ int fail = 0;
+
+ INIT_LIST_HEAD(&cull);
+
+ /* NB: use lnet_net_lock(0) to serialize operations on test peers */
+ lnet_net_lock(0);
+
+ list_for_each_safe(el, next, &the_lnet.ln_test_peers) {
+ tp = list_entry(el, lnet_test_peer_t, tp_list);
+
+ if (tp->tp_threshold == 0) {
+ /* zombie entry */
+ if (outgoing) {
+ /* only cull zombies on outgoing tests,
+ * since we may be at interrupt priority on
+ * incoming messages. */
+ list_del(&tp->tp_list);
+ list_add(&tp->tp_list, &cull);
+ }
+ continue;
+ }
- if (tp->tp_nid == LNET_NID_ANY || /* fail every peer */
- nid == tp->tp_nid) { /* fail this peer */
- fail = 1;
-
- if (tp->tp_threshold != LNET_MD_THRESH_INF) {
- tp->tp_threshold--;
- if (outgoing &&
- tp->tp_threshold == 0) {
- /* see above */
- cfs_list_del (&tp->tp_list);
- cfs_list_add (&tp->tp_list, &cull);
- }
- }
- break;
- }
- }
+ if (tp->tp_nid == LNET_NID_ANY || /* fail every peer */
+ nid == tp->tp_nid) { /* fail this peer */
+ fail = 1;
+
+ if (tp->tp_threshold != LNET_MD_THRESH_INF) {
+ tp->tp_threshold--;
+ if (outgoing &&
+ tp->tp_threshold == 0) {
+ /* see above */
+ list_del(&tp->tp_list);
+ list_add(&tp->tp_list, &cull);
+ }
+ }
+ break;
+ }
+ }
- LNET_UNLOCK ();
+ lnet_net_unlock(0);
- while (!cfs_list_empty (&cull)) {
- tp = cfs_list_entry (cull.next, lnet_test_peer_t, tp_list);
- cfs_list_del (&tp->tp_list);
+ while (!list_empty(&cull)) {
+ tp = list_entry(cull.next, lnet_test_peer_t, tp_list);
+ list_del(&tp->tp_list);
- LIBCFS_FREE(tp, sizeof (*tp));
- }
+ LIBCFS_FREE(tp, sizeof(*tp));
+ }
- return (fail);
+ return fail;
}
unsigned int
-lnet_iov_nob (unsigned int niov, struct iovec *iov)
+lnet_iov_nob(unsigned int niov, struct kvec *iov)
{
unsigned int nob = 0;
+ LASSERT(niov == 0 || iov != NULL);
while (niov-- > 0)
nob += (iov++)->iov_len;
return (nob);
}
+EXPORT_SYMBOL(lnet_iov_nob);
void
-lnet_copy_iov2iov (unsigned int ndiov, struct iovec *diov, unsigned int doffset,
- unsigned int nsiov, struct iovec *siov, unsigned int soffset,
- unsigned int nob)
+lnet_copy_iov2iov(unsigned int ndiov, struct kvec *diov, unsigned int doffset,
+ unsigned int nsiov, struct kvec *siov, unsigned int soffset,
+ unsigned int nob)
{
/* NB diov, siov are READ-ONLY */
unsigned int this_nob;
}
} while (nob > 0);
}
+EXPORT_SYMBOL(lnet_copy_iov2iov);
int
-lnet_extract_iov (int dst_niov, struct iovec *dst,
- int src_niov, struct iovec *src,
- unsigned int offset, unsigned int len)
+lnet_extract_iov(int dst_niov, struct kvec *dst,
+ int src_niov, struct kvec *src,
+ unsigned int offset, unsigned int len)
{
/* Initialise 'dst' to the subset of 'src' starting at 'offset',
* for exactly 'len' bytes, and return the number of entries.
offset = 0;
}
}
+EXPORT_SYMBOL(lnet_extract_iov);
-#ifndef __KERNEL__
-unsigned int
-lnet_kiov_nob (unsigned int niov, lnet_kiov_t *kiov)
-{
- LASSERT (0);
- return (0);
-}
-
-void
-lnet_copy_kiov2kiov (unsigned int ndkiov, lnet_kiov_t *dkiov, unsigned int doffset,
- unsigned int nskiov, lnet_kiov_t *skiov, unsigned int soffset,
- unsigned int nob)
-{
- LASSERT (0);
-}
-
-void
-lnet_copy_kiov2iov (unsigned int niov, struct iovec *iov, unsigned int iovoffset,
- unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
- unsigned int nob)
-{
- LASSERT (0);
-}
-
-void
-lnet_copy_iov2kiov (unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
- unsigned int niov, struct iovec *iov, unsigned int iovoffset,
- unsigned int nob)
-{
- LASSERT (0);
-}
-
-int
-lnet_extract_kiov (int dst_niov, lnet_kiov_t *dst,
- int src_niov, lnet_kiov_t *src,
- unsigned int offset, unsigned int len)
-{
- LASSERT (0);
-}
-
-#else /* __KERNEL__ */
unsigned int
-lnet_kiov_nob (unsigned int niov, lnet_kiov_t *kiov)
+lnet_kiov_nob(unsigned int niov, lnet_kiov_t *kiov)
{
unsigned int nob = 0;
+ LASSERT(niov == 0 || kiov != NULL);
while (niov-- > 0)
nob += (kiov++)->kiov_len;
return (nob);
}
+EXPORT_SYMBOL(lnet_kiov_nob);
void
-lnet_copy_kiov2kiov (unsigned int ndiov, lnet_kiov_t *diov, unsigned int doffset,
- unsigned int nsiov, lnet_kiov_t *siov, unsigned int soffset,
- unsigned int nob)
+lnet_copy_kiov2kiov(unsigned int ndiov, lnet_kiov_t *diov, unsigned int doffset,
+ unsigned int nsiov, lnet_kiov_t *siov, unsigned int soffset,
+ unsigned int nob)
{
- /* NB diov, siov are READ-ONLY */
- unsigned int this_nob;
- char *daddr = NULL;
- char *saddr = NULL;
+ /* NB diov, siov are READ-ONLY */
+ unsigned int this_nob;
+ char *daddr = NULL;
+ char *saddr = NULL;
- if (nob == 0)
- return;
+ if (nob == 0)
+ return;
- LASSERT (!cfs_in_interrupt ());
+ LASSERT (!in_interrupt ());
- LASSERT (ndiov > 0);
+ LASSERT (ndiov > 0);
while (doffset >= diov->kiov_len) {
doffset -= diov->kiov_len;
diov++;
siov->kiov_len - soffset);
this_nob = MIN(this_nob, nob);
- if (daddr == NULL)
- daddr = ((char *)cfs_kmap(diov->kiov_page)) +
- diov->kiov_offset + doffset;
- if (saddr == NULL)
- saddr = ((char *)cfs_kmap(siov->kiov_page)) +
- siov->kiov_offset + soffset;
+ if (daddr == NULL)
+ daddr = ((char *)kmap(diov->kiov_page)) +
+ diov->kiov_offset + doffset;
+ if (saddr == NULL)
+ saddr = ((char *)kmap(siov->kiov_page)) +
+ siov->kiov_offset + soffset;
- /* Vanishing risk of kmap deadlock when mapping 2 pages.
- * However in practice at least one of the kiovs will be mapped
- * kernel pages and the map/unmap will be NOOPs */
+ /* Vanishing risk of kmap deadlock when mapping 2 pages.
+ * However in practice at least one of the kiovs will be mapped
+ * kernel pages and the map/unmap will be NOOPs */
- memcpy (daddr, saddr, this_nob);
- nob -= this_nob;
+ memcpy (daddr, saddr, this_nob);
+ nob -= this_nob;
- if (diov->kiov_len > doffset + this_nob) {
- daddr += this_nob;
- doffset += this_nob;
- } else {
- cfs_kunmap(diov->kiov_page);
- daddr = NULL;
- diov++;
- ndiov--;
- doffset = 0;
- }
+ if (diov->kiov_len > doffset + this_nob) {
+ daddr += this_nob;
+ doffset += this_nob;
+ } else {
+ kunmap(diov->kiov_page);
+ daddr = NULL;
+ diov++;
+ ndiov--;
+ doffset = 0;
+ }
- if (siov->kiov_len > soffset + this_nob) {
- saddr += this_nob;
- soffset += this_nob;
- } else {
- cfs_kunmap(siov->kiov_page);
- saddr = NULL;
- siov++;
- nsiov--;
- soffset = 0;
- }
- } while (nob > 0);
+ if (siov->kiov_len > soffset + this_nob) {
+ saddr += this_nob;
+ soffset += this_nob;
+ } else {
+ kunmap(siov->kiov_page);
+ saddr = NULL;
+ siov++;
+ nsiov--;
+ soffset = 0;
+ }
+ } while (nob > 0);
- if (daddr != NULL)
- cfs_kunmap(diov->kiov_page);
- if (saddr != NULL)
- cfs_kunmap(siov->kiov_page);
+ if (daddr != NULL)
+ kunmap(diov->kiov_page);
+ if (saddr != NULL)
+ kunmap(siov->kiov_page);
}
+EXPORT_SYMBOL(lnet_copy_kiov2kiov);
void
-lnet_copy_kiov2iov (unsigned int niov, struct iovec *iov, unsigned int iovoffset,
+lnet_copy_kiov2iov (unsigned int niov, struct kvec *iov, unsigned int iovoffset,
unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
unsigned int nob)
{
- /* NB iov, kiov are READ-ONLY */
- unsigned int this_nob;
- char *addr = NULL;
+ /* NB iov, kiov are READ-ONLY */
+ unsigned int this_nob;
+ char *addr = NULL;
- if (nob == 0)
- return;
+ if (nob == 0)
+ return;
- LASSERT (!cfs_in_interrupt ());
+ LASSERT (!in_interrupt ());
- LASSERT (niov > 0);
+ LASSERT (niov > 0);
while (iovoffset >= iov->iov_len) {
iovoffset -= iov->iov_len;
iov++;
this_nob = MIN(this_nob, nob);
if (addr == NULL)
- addr = ((char *)cfs_kmap(kiov->kiov_page)) +
+ addr = ((char *)kmap(kiov->kiov_page)) +
kiov->kiov_offset + kiovoffset;
memcpy ((char *)iov->iov_base + iovoffset, addr, this_nob);
addr += this_nob;
kiovoffset += this_nob;
} else {
- cfs_kunmap(kiov->kiov_page);
- addr = NULL;
- kiov++;
- nkiov--;
- kiovoffset = 0;
- }
+ kunmap(kiov->kiov_page);
+ addr = NULL;
+ kiov++;
+ nkiov--;
+ kiovoffset = 0;
+ }
- } while (nob > 0);
+ } while (nob > 0);
- if (addr != NULL)
- cfs_kunmap(kiov->kiov_page);
+ if (addr != NULL)
+ kunmap(kiov->kiov_page);
}
+EXPORT_SYMBOL(lnet_copy_kiov2iov);
void
-lnet_copy_iov2kiov (unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
- unsigned int niov, struct iovec *iov, unsigned int iovoffset,
- unsigned int nob)
+lnet_copy_iov2kiov(unsigned int nkiov, lnet_kiov_t *kiov, unsigned int kiovoffset,
+ unsigned int niov, struct kvec *iov, unsigned int iovoffset,
+ unsigned int nob)
{
- /* NB kiov, iov are READ-ONLY */
- unsigned int this_nob;
- char *addr = NULL;
+ /* NB kiov, iov are READ-ONLY */
+ unsigned int this_nob;
+ char *addr = NULL;
- if (nob == 0)
- return;
+ if (nob == 0)
+ return;
- LASSERT (!cfs_in_interrupt ());
+ LASSERT (!in_interrupt ());
- LASSERT (nkiov > 0);
+ LASSERT (nkiov > 0);
while (kiovoffset >= kiov->kiov_len) {
kiovoffset -= kiov->kiov_len;
kiov++;
this_nob = MIN(this_nob, nob);
if (addr == NULL)
- addr = ((char *)cfs_kmap(kiov->kiov_page)) +
- kiov->kiov_offset + kiovoffset;
+ addr = ((char *)kmap(kiov->kiov_page)) +
+ kiov->kiov_offset + kiovoffset;
- memcpy (addr, (char *)iov->iov_base + iovoffset, this_nob);
- nob -= this_nob;
+ memcpy (addr, (char *)iov->iov_base + iovoffset, this_nob);
+ nob -= this_nob;
- if (kiov->kiov_len > kiovoffset + this_nob) {
- addr += this_nob;
- kiovoffset += this_nob;
- } else {
- cfs_kunmap(kiov->kiov_page);
- addr = NULL;
- kiov++;
- nkiov--;
- kiovoffset = 0;
- }
+ if (kiov->kiov_len > kiovoffset + this_nob) {
+ addr += this_nob;
+ kiovoffset += this_nob;
+ } else {
+ kunmap(kiov->kiov_page);
+ addr = NULL;
+ kiov++;
+ nkiov--;
+ kiovoffset = 0;
+ }
- if (iov->iov_len > iovoffset + this_nob) {
- iovoffset += this_nob;
- } else {
- iov++;
- niov--;
- iovoffset = 0;
- }
- } while (nob > 0);
+ if (iov->iov_len > iovoffset + this_nob) {
+ iovoffset += this_nob;
+ } else {
+ iov++;
+ niov--;
+ iovoffset = 0;
+ }
+ } while (nob > 0);
- if (addr != NULL)
- cfs_kunmap(kiov->kiov_page);
+ if (addr != NULL)
+ kunmap(kiov->kiov_page);
}
+EXPORT_SYMBOL(lnet_copy_iov2kiov);
int
-lnet_extract_kiov (int dst_niov, lnet_kiov_t *dst,
- int src_niov, lnet_kiov_t *src,
- unsigned int offset, unsigned int len)
+lnet_extract_kiov(int dst_niov, lnet_kiov_t *dst,
+ int src_niov, lnet_kiov_t *src,
+ unsigned int offset, unsigned int len)
{
/* Initialise 'dst' to the subset of 'src' starting at 'offset',
* for exactly 'len' bytes, and return the number of entries.
dst->kiov_page = src->kiov_page;
dst->kiov_offset = src->kiov_offset + offset;
- if (len <= frag_len) {
- dst->kiov_len = len;
- LASSERT (dst->kiov_offset + dst->kiov_len <= CFS_PAGE_SIZE);
- return (niov);
- }
+ if (len <= frag_len) {
+ dst->kiov_len = len;
+ LASSERT (dst->kiov_offset + dst->kiov_len <= PAGE_CACHE_SIZE);
+ return niov;
+ }
- dst->kiov_len = frag_len;
- LASSERT (dst->kiov_offset + dst->kiov_len <= CFS_PAGE_SIZE);
+ dst->kiov_len = frag_len;
+ LASSERT (dst->kiov_offset + dst->kiov_len <= PAGE_CACHE_SIZE);
len -= frag_len;
dst++;
offset = 0;
}
}
-#endif
+EXPORT_SYMBOL(lnet_extract_kiov);
void
lnet_ni_recv(lnet_ni_t *ni, void *private, lnet_msg_t *msg, int delayed,
unsigned int offset, unsigned int mlen, unsigned int rlen)
{
- unsigned int niov = 0;
- struct iovec *iov = NULL;
- lnet_kiov_t *kiov = NULL;
- int rc;
+ unsigned int niov = 0;
+ struct kvec *iov = NULL;
+ lnet_kiov_t *kiov = NULL;
+ int rc;
- LASSERT (!cfs_in_interrupt ());
- LASSERT (mlen == 0 || msg != NULL);
+ LASSERT (!in_interrupt ());
+ LASSERT (mlen == 0 || msg != NULL);
if (msg != NULL) {
LASSERT(msg->msg_receiving);
lnet_finalize(ni, msg, rc);
}
-void
+static void
lnet_setpayloadbuffer(lnet_msg_t *msg)
{
lnet_libmd_t *md = msg->msg_md;
void
lnet_prep_send(lnet_msg_t *msg, int type, lnet_process_id_t target,
- unsigned int offset, unsigned int len)
+ unsigned int offset, unsigned int len)
{
msg->msg_type = type;
msg->msg_target = target;
msg->msg_hdr.payload_length = cpu_to_le32(len);
}
-void
+static void
lnet_ni_send(lnet_ni_t *ni, lnet_msg_t *msg)
{
- void *priv = msg->msg_private;
- int rc;
+ void *priv = msg->msg_private;
+ int rc;
- LASSERT (!cfs_in_interrupt ());
- LASSERT (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND ||
- (msg->msg_txcredit && msg->msg_peertxcredit));
+ LASSERT (!in_interrupt ());
+ LASSERT (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND ||
+ (msg->msg_txcredit && msg->msg_peertxcredit));
- rc = (ni->ni_lnd->lnd_send)(ni, priv, msg);
- if (rc < 0)
- lnet_finalize(ni, msg, rc);
+ rc = (ni->ni_lnd->lnd_send)(ni, priv, msg);
+ if (rc < 0)
+ lnet_finalize(ni, msg, rc);
}
-int
+static int
lnet_ni_eager_recv(lnet_ni_t *ni, lnet_msg_t *msg)
{
int rc;
LASSERT(!msg->msg_sending);
LASSERT(msg->msg_receiving);
+ LASSERT(!msg->msg_rx_ready_delay);
LASSERT(ni->ni_lnd->lnd_eager_recv != NULL);
msg->msg_rx_ready_delay = 1;
return rc;
}
-/* NB: caller shall hold a ref on 'lp' as I'd drop LNET_LOCK */
-void
+/* NB: caller shall hold a ref on 'lp' as I'd drop lnet_net_lock */
+static void
lnet_ni_query_locked(lnet_ni_t *ni, lnet_peer_t *lp)
{
- cfs_time_t last_alive = 0;
+ cfs_time_t last_alive = 0;
- LASSERT (lnet_peer_aliveness_enabled(lp));
- LASSERT (ni->ni_lnd->lnd_query != NULL);
- LASSERT (the_lnet.ln_routing == 1);
+ LASSERT(lnet_peer_aliveness_enabled(lp));
+ LASSERT(ni->ni_lnd->lnd_query != NULL);
- LNET_UNLOCK();
- (ni->ni_lnd->lnd_query)(ni, lp->lp_nid, &last_alive);
- LNET_LOCK();
+ lnet_net_unlock(lp->lp_cpt);
+ (ni->ni_lnd->lnd_query)(ni, lp->lp_nid, &last_alive);
+ lnet_net_lock(lp->lp_cpt);
- lp->lp_last_query = cfs_time_current();
+ lp->lp_last_query = cfs_time_current();
- if (last_alive != 0) /* NI has updated timestamp */
- lp->lp_last_alive = last_alive;
+ if (last_alive != 0) /* NI has updated timestamp */
+ lp->lp_last_alive = last_alive;
}
-/* NB: always called with LNET_LOCK held */
+/* NB: always called with lnet_net_lock held */
static inline int
lnet_peer_is_alive (lnet_peer_t *lp, cfs_time_t now)
{
cfs_time_t deadline;
LASSERT (lnet_peer_aliveness_enabled(lp));
- LASSERT (the_lnet.ln_routing == 1);
/* Trust lnet_notify() if it has more recent aliveness news, but
* ignore the initial assumed death (see lnet_peers_start_down()).
/* NB: returns 1 when alive, 0 when dead, negative when error;
- * may drop the LNET_LOCK */
-int
+ * may drop the lnet_net_lock */
+static int
lnet_peer_alive_locked (lnet_peer_t *lp)
{
cfs_time_t now = cfs_time_current();
- /* LU-630: only router checks peer health. */
- if (the_lnet.ln_routing == 0)
- return 1;
-
if (!lnet_peer_aliveness_enabled(lp))
return -ENODEV;
return 0;
}
-int
-lnet_post_send_locked (lnet_msg_t *msg, int do_send)
+/**
+ * \param msg The message to be sent.
+ * \param do_send True if lnet_ni_send() should be called in this function.
+ * lnet_send() is going to lnet_net_unlock immediately after this, so
+ * it sets do_send FALSE and I don't do the unlock/send/lock bit.
+ *
+ * \retval LNET_CREDIT_OK If \a msg sent or OK to send.
+ * \retval LNET_CREDIT_WAIT If \a msg blocked for credit.
+ * \retval -EHOSTUNREACH If the next hop of the message appears dead.
+ * \retval -ECANCELED If the MD of the message has been unlinked.
+ */
+static int
+lnet_post_send_locked(lnet_msg_t *msg, int do_send)
{
- /* lnet_send is going to LNET_UNLOCK immediately after this, so it sets
- * do_send FALSE and I don't do the unlock/send/lock bit. I return
- * EAGAIN if msg blocked, EHOSTUNREACH if msg_txpeer appears dead, and
- * 0 if sent or OK to send */
- lnet_peer_t *lp = msg->msg_txpeer;
- lnet_ni_t *ni = lp->lp_ni;
-
- /* non-lnet_send() callers have checked before */
+ lnet_peer_t *lp = msg->msg_txpeer;
+ lnet_ni_t *ni = lp->lp_ni;
+ int cpt = msg->msg_tx_cpt;
+ struct lnet_tx_queue *tq = ni->ni_tx_queues[cpt];
+
+ /* non-lnet_send() callers have checked before */
LASSERT(!do_send || msg->msg_tx_delayed);
LASSERT(!msg->msg_receiving);
+ LASSERT(msg->msg_tx_committed);
/* NB 'lp' is always the next hop */
if ((msg->msg_target.pid & LNET_PID_USERFLAG) == 0 &&
lnet_peer_alive_locked(lp) == 0) {
- the_lnet.ln_counters->drop_count++;
- the_lnet.ln_counters->drop_length += msg->msg_len;
- LNET_UNLOCK();
+ the_lnet.ln_counters[cpt]->drop_count++;
+ the_lnet.ln_counters[cpt]->drop_length += msg->msg_len;
+ lnet_net_unlock(cpt);
- CNETERR("Dropping message for %s: peer not alive\n",
- libcfs_id2str(msg->msg_target));
- if (do_send)
- lnet_finalize(ni, msg, -EHOSTUNREACH);
+ CNETERR("Dropping message for %s: peer not alive\n",
+ libcfs_id2str(msg->msg_target));
+ if (do_send)
+ lnet_finalize(ni, msg, -EHOSTUNREACH);
- LNET_LOCK();
- return EHOSTUNREACH;
- }
+ lnet_net_lock(cpt);
+ return -EHOSTUNREACH;
+ }
+
+ if (msg->msg_md != NULL &&
+ (msg->msg_md->md_flags & LNET_MD_FLAG_ABORTED) != 0) {
+ lnet_net_unlock(cpt);
+
+ CNETERR("Aborting message for %s: LNetM[DE]Unlink() already "
+ "called on the MD/ME.\n",
+ libcfs_id2str(msg->msg_target));
+ if (do_send)
+ lnet_finalize(ni, msg, -ECANCELED);
+
+ lnet_net_lock(cpt);
+ return -ECANCELED;
+ }
- if (!msg->msg_peertxcredit) {
- LASSERT ((lp->lp_txcredits < 0) ==
- !cfs_list_empty(&lp->lp_txq));
+ if (!msg->msg_peertxcredit) {
+ LASSERT((lp->lp_txcredits < 0) ==
+ !list_empty(&lp->lp_txq));
- msg->msg_peertxcredit = 1;
- lp->lp_txqnob += msg->msg_len + sizeof(lnet_hdr_t);
- lp->lp_txcredits--;
+ msg->msg_peertxcredit = 1;
+ lp->lp_txqnob += msg->msg_len + sizeof(lnet_hdr_t);
+ lp->lp_txcredits--;
- if (lp->lp_txcredits < lp->lp_mintxcredits)
- lp->lp_mintxcredits = lp->lp_txcredits;
+ if (lp->lp_txcredits < lp->lp_mintxcredits)
+ lp->lp_mintxcredits = lp->lp_txcredits;
- if (lp->lp_txcredits < 0) {
+ if (lp->lp_txcredits < 0) {
msg->msg_tx_delayed = 1;
- cfs_list_add_tail(&msg->msg_list, &lp->lp_txq);
- return EAGAIN;
- }
- }
+ list_add_tail(&msg->msg_list, &lp->lp_txq);
+ return LNET_CREDIT_WAIT;
+ }
+ }
- if (!msg->msg_txcredit) {
- LASSERT ((ni->ni_txcredits < 0) ==
- !cfs_list_empty(&ni->ni_txq));
+ if (!msg->msg_txcredit) {
+ LASSERT((tq->tq_credits < 0) ==
+ !list_empty(&tq->tq_delayed));
- msg->msg_txcredit = 1;
- ni->ni_txcredits--;
+ msg->msg_txcredit = 1;
+ tq->tq_credits--;
- if (ni->ni_txcredits < ni->ni_mintxcredits)
- ni->ni_mintxcredits = ni->ni_txcredits;
+ if (tq->tq_credits < tq->tq_credits_min)
+ tq->tq_credits_min = tq->tq_credits;
- if (ni->ni_txcredits < 0) {
+ if (tq->tq_credits < 0) {
msg->msg_tx_delayed = 1;
- cfs_list_add_tail(&msg->msg_list, &ni->ni_txq);
- return EAGAIN;
- }
- }
+ list_add_tail(&msg->msg_list, &tq->tq_delayed);
+ return LNET_CREDIT_WAIT;
+ }
+ }
- if (do_send) {
- LNET_UNLOCK();
- lnet_ni_send(ni, msg);
- LNET_LOCK();
- }
- return 0;
+ if (do_send) {
+ lnet_net_unlock(cpt);
+ lnet_ni_send(ni, msg);
+ lnet_net_lock(cpt);
+ }
+ return LNET_CREDIT_OK;
}
-#ifdef __KERNEL__
-lnet_rtrbufpool_t *
+static lnet_rtrbufpool_t *
lnet_msg2bufpool(lnet_msg_t *msg)
{
- lnet_rtrbufpool_t *rbp = &the_lnet.ln_rtrpools[0];
+ lnet_rtrbufpool_t *rbp;
+ int cpt;
- LASSERT (msg->msg_len <= LNET_MTU);
- while (msg->msg_len > (unsigned int)rbp->rbp_npages * CFS_PAGE_SIZE) {
- rbp++;
- LASSERT (rbp < &the_lnet.ln_rtrpools[LNET_NRBPOOLS]);
- }
+ LASSERT(msg->msg_rx_committed);
+
+ cpt = msg->msg_rx_cpt;
+ rbp = &the_lnet.ln_rtrpools[cpt][0];
+
+ LASSERT(msg->msg_len <= LNET_MTU);
+ while (msg->msg_len > (unsigned int)rbp->rbp_npages * PAGE_CACHE_SIZE) {
+ rbp++;
+ LASSERT(rbp < &the_lnet.ln_rtrpools[cpt][LNET_NRBPOOLS]);
+ }
- return rbp;
+ return rbp;
}
-int
+static int
lnet_post_routed_recv_locked (lnet_msg_t *msg, int do_recv)
{
- /* lnet_parse is going to LNET_UNLOCK immediately after this, so it
- * sets do_recv FALSE and I don't do the unlock/send/lock bit. I
- * return EAGAIN if msg blocked and 0 if received or OK to receive */
+ /* lnet_parse is going to lnet_net_unlock immediately after this, so it
+ * sets do_recv FALSE and I don't do the unlock/send/lock bit.
+ * I return LNET_CREDIT_WAIT if msg blocked and LNET_CREDIT_OK if
+ * received or OK to receive */
lnet_peer_t *lp = msg->msg_rxpeer;
lnet_rtrbufpool_t *rbp;
lnet_rtrbuf_t *rb;
/* non-lnet_parse callers only receive delayed messages */
LASSERT(!do_recv || msg->msg_rx_delayed);
- if (!msg->msg_peerrtrcredit) {
- LASSERT ((lp->lp_rtrcredits < 0) ==
- !cfs_list_empty(&lp->lp_rtrq));
+ if (!msg->msg_peerrtrcredit) {
+ LASSERT((lp->lp_rtrcredits < 0) ==
+ !list_empty(&lp->lp_rtrq));
msg->msg_peerrtrcredit = 1;
lp->lp_rtrcredits--;
if (lp->lp_rtrcredits < lp->lp_minrtrcredits)
lp->lp_minrtrcredits = lp->lp_rtrcredits;
- if (lp->lp_rtrcredits < 0) {
- /* must have checked eager_recv before here */
+ if (lp->lp_rtrcredits < 0) {
+ /* must have checked eager_recv before here */
LASSERT(msg->msg_rx_ready_delay);
msg->msg_rx_delayed = 1;
- cfs_list_add_tail(&msg->msg_list, &lp->lp_rtrq);
- return EAGAIN;
- }
- }
-
- rbp = lnet_msg2bufpool(msg);
+ list_add_tail(&msg->msg_list, &lp->lp_rtrq);
+ return LNET_CREDIT_WAIT;
+ }
+ }
- if (!msg->msg_rtrcredit) {
- LASSERT ((rbp->rbp_credits < 0) ==
- !cfs_list_empty(&rbp->rbp_msgs));
+ rbp = lnet_msg2bufpool(msg);
- msg->msg_rtrcredit = 1;
- rbp->rbp_credits--;
- if (rbp->rbp_credits < rbp->rbp_mincredits)
- rbp->rbp_mincredits = rbp->rbp_credits;
+ if (!msg->msg_rtrcredit) {
+ msg->msg_rtrcredit = 1;
+ rbp->rbp_credits--;
+ if (rbp->rbp_credits < rbp->rbp_mincredits)
+ rbp->rbp_mincredits = rbp->rbp_credits;
- if (rbp->rbp_credits < 0) {
- /* must have checked eager_recv before here */
+ if (rbp->rbp_credits < 0) {
+ /* must have checked eager_recv before here */
LASSERT(msg->msg_rx_ready_delay);
msg->msg_rx_delayed = 1;
- cfs_list_add_tail(&msg->msg_list, &rbp->rbp_msgs);
- return EAGAIN;
- }
- }
+ list_add_tail(&msg->msg_list, &rbp->rbp_msgs);
+ return LNET_CREDIT_WAIT;
+ }
+ }
- LASSERT (!cfs_list_empty(&rbp->rbp_bufs));
- rb = cfs_list_entry(rbp->rbp_bufs.next, lnet_rtrbuf_t, rb_list);
- cfs_list_del(&rb->rb_list);
+ LASSERT(!list_empty(&rbp->rbp_bufs));
+ rb = list_entry(rbp->rbp_bufs.next, lnet_rtrbuf_t, rb_list);
+ list_del(&rb->rb_list);
msg->msg_niov = rbp->rbp_npages;
msg->msg_kiov = &rb->rb_kiov[0];
if (do_recv) {
- LNET_UNLOCK();
- lnet_ni_recv(lp->lp_ni, msg->msg_private, msg, 1,
- 0, msg->msg_len, msg->msg_len);
- LNET_LOCK();
- }
- return 0;
+ int cpt = msg->msg_rx_cpt;
+
+ lnet_net_unlock(cpt);
+ lnet_ni_recv(lp->lp_ni, msg->msg_private, msg, 1,
+ 0, msg->msg_len, msg->msg_len);
+ lnet_net_lock(cpt);
+ }
+ return LNET_CREDIT_OK;
}
-#endif
void
lnet_return_tx_credits_locked(lnet_msg_t *msg)
{
lnet_peer_t *txpeer = msg->msg_txpeer;
lnet_msg_t *msg2;
- lnet_ni_t *ni;
- if (msg->msg_txcredit) {
- /* give back NI txcredits */
- msg->msg_txcredit = 0;
- ni = txpeer->lp_ni;
+ if (msg->msg_txcredit) {
+ struct lnet_ni *ni = txpeer->lp_ni;
+ struct lnet_tx_queue *tq = ni->ni_tx_queues[msg->msg_tx_cpt];
- LASSERT((ni->ni_txcredits < 0) == !cfs_list_empty(&ni->ni_txq));
+ /* give back NI txcredits */
+ msg->msg_txcredit = 0;
- ni->ni_txcredits++;
- if (ni->ni_txcredits <= 0) {
- msg2 = cfs_list_entry(ni->ni_txq.next, lnet_msg_t,
- msg_list);
- cfs_list_del(&msg2->msg_list);
+ LASSERT((tq->tq_credits < 0) ==
+ !list_empty(&tq->tq_delayed));
- LASSERT(msg2->msg_txpeer->lp_ni == ni);
+ tq->tq_credits++;
+ if (tq->tq_credits <= 0) {
+ msg2 = list_entry(tq->tq_delayed.next,
+ lnet_msg_t, msg_list);
+ list_del(&msg2->msg_list);
+
+ LASSERT(msg2->msg_txpeer->lp_ni == ni);
LASSERT(msg2->msg_tx_delayed);
(void) lnet_post_send_locked(msg2, 1);
msg->msg_peertxcredit = 0;
LASSERT((txpeer->lp_txcredits < 0) ==
- !cfs_list_empty(&txpeer->lp_txq));
+ !list_empty(&txpeer->lp_txq));
txpeer->lp_txqnob -= msg->msg_len + sizeof(lnet_hdr_t);
LASSERT (txpeer->lp_txqnob >= 0);
txpeer->lp_txcredits++;
if (txpeer->lp_txcredits <= 0) {
- msg2 = cfs_list_entry(txpeer->lp_txq.next,
+ msg2 = list_entry(txpeer->lp_txq.next,
lnet_msg_t, msg_list);
- cfs_list_del(&msg2->msg_list);
+ list_del(&msg2->msg_list);
LASSERT(msg2->msg_txpeer == txpeer);
LASSERT(msg2->msg_tx_delayed);
}
void
+lnet_schedule_blocked_locked(lnet_rtrbufpool_t *rbp)
+{
+ lnet_msg_t *msg;
+
+ if (list_empty(&rbp->rbp_msgs))
+ return;
+ msg = list_entry(rbp->rbp_msgs.next,
+ lnet_msg_t, msg_list);
+ list_del(&msg->msg_list);
+
+ (void)lnet_post_routed_recv_locked(msg, 1);
+}
+
+void
+lnet_drop_routed_msgs_locked(struct list_head *list, int cpt)
+{
+ lnet_msg_t *msg;
+ lnet_msg_t *tmp;
+ struct list_head drop;
+
+ INIT_LIST_HEAD(&drop);
+
+ list_splice_init(list, &drop);
+
+ lnet_net_unlock(cpt);
+
+ list_for_each_entry_safe(msg, tmp, &drop, msg_list) {
+ lnet_ni_recv(msg->msg_rxpeer->lp_ni, msg->msg_private, NULL,
+ 0, 0, 0, msg->msg_hdr.payload_length);
+ list_del_init(&msg->msg_list);
+ lnet_finalize(NULL, msg, -ECANCELED);
+ }
+
+ lnet_net_lock(cpt);
+}
+
+void
lnet_return_rx_credits_locked(lnet_msg_t *msg)
{
lnet_peer_t *rxpeer = msg->msg_rxpeer;
-#ifdef __KERNEL__
lnet_msg_t *msg2;
- if (msg->msg_rtrcredit) {
- /* give back global router credits */
- lnet_rtrbuf_t *rb;
- lnet_rtrbufpool_t *rbp;
+ if (msg->msg_rtrcredit) {
+ /* give back global router credits */
+ lnet_rtrbuf_t *rb;
+ lnet_rtrbufpool_t *rbp;
- /* NB If a msg ever blocks for a buffer in rbp_msgs, it stays
- * there until it gets one allocated, or aborts the wait
- * itself */
- LASSERT (msg->msg_kiov != NULL);
+ /* NB If a msg ever blocks for a buffer in rbp_msgs, it stays
+ * there until it gets one allocated, or aborts the wait
+ * itself */
+ LASSERT(msg->msg_kiov != NULL);
- rb = cfs_list_entry(msg->msg_kiov, lnet_rtrbuf_t, rb_kiov[0]);
- rbp = rb->rb_pool;
- LASSERT (rbp == lnet_msg2bufpool(msg));
+ rb = list_entry(msg->msg_kiov, lnet_rtrbuf_t, rb_kiov[0]);
+ rbp = rb->rb_pool;
- msg->msg_kiov = NULL;
- msg->msg_rtrcredit = 0;
+ msg->msg_kiov = NULL;
+ msg->msg_rtrcredit = 0;
- LASSERT((rbp->rbp_credits < 0) ==
- !cfs_list_empty(&rbp->rbp_msgs));
- LASSERT((rbp->rbp_credits > 0) ==
- !cfs_list_empty(&rbp->rbp_bufs));
+ LASSERT(rbp == lnet_msg2bufpool(msg));
- cfs_list_add(&rb->rb_list, &rbp->rbp_bufs);
- rbp->rbp_credits++;
- if (rbp->rbp_credits <= 0) {
- msg2 = cfs_list_entry(rbp->rbp_msgs.next,
- lnet_msg_t, msg_list);
- cfs_list_del(&msg2->msg_list);
+ LASSERT((rbp->rbp_credits > 0) ==
+ !list_empty(&rbp->rbp_bufs));
- (void) lnet_post_routed_recv_locked(msg2, 1);
- }
- }
+ /* If routing is now turned off, we just drop this buffer and
+ * don't bother trying to return credits. */
+ if (!the_lnet.ln_routing) {
+ lnet_destroy_rtrbuf(rb, rbp->rbp_npages);
+ goto routing_off;
+ }
+
+ /* It is possible that a user has lowered the desired number of
+ * buffers in this pool. Make sure we never put back
+ * more buffers than the stated number. */
+ if (unlikely(rbp->rbp_credits >= rbp->rbp_req_nbuffers)) {
+ /* Discard this buffer so we don't have too
+ * many. */
+ lnet_destroy_rtrbuf(rb, rbp->rbp_npages);
+ rbp->rbp_nbuffers--;
+ } else {
+ list_add(&rb->rb_list, &rbp->rbp_bufs);
+ rbp->rbp_credits++;
+ if (rbp->rbp_credits <= 0)
+ lnet_schedule_blocked_locked(rbp);
+ }
+ }
- if (msg->msg_peerrtrcredit) {
- /* give back peer router credits */
- msg->msg_peerrtrcredit = 0;
+routing_off:
+ if (msg->msg_peerrtrcredit) {
+ /* give back peer router credits */
+ msg->msg_peerrtrcredit = 0;
- LASSERT((rxpeer->lp_rtrcredits < 0) ==
- !cfs_list_empty(&rxpeer->lp_rtrq));
+ LASSERT((rxpeer->lp_rtrcredits < 0) ==
+ !list_empty(&rxpeer->lp_rtrq));
- rxpeer->lp_rtrcredits++;
- if (rxpeer->lp_rtrcredits <= 0) {
- msg2 = cfs_list_entry(rxpeer->lp_rtrq.next,
- lnet_msg_t, msg_list);
- cfs_list_del(&msg2->msg_list);
+ rxpeer->lp_rtrcredits++;
- (void) lnet_post_routed_recv_locked(msg2, 1);
- }
- }
-#else
- LASSERT (!msg->msg_rtrcredit);
- LASSERT (!msg->msg_peerrtrcredit);
-#endif
- if (rxpeer != NULL) {
- msg->msg_rxpeer = NULL;
- lnet_peer_decref_locked(rxpeer);
- }
+ /* drop all messages which are queued to be routed on that
+ * peer. */
+ if (!the_lnet.ln_routing) {
+ lnet_drop_routed_msgs_locked(&rxpeer->lp_rtrq,
+ msg->msg_rx_cpt);
+ } else if (rxpeer->lp_rtrcredits <= 0) {
+ msg2 = list_entry(rxpeer->lp_rtrq.next,
+ lnet_msg_t, msg_list);
+ list_del(&msg2->msg_list);
+
+ (void) lnet_post_routed_recv_locked(msg2, 1);
+ }
+ }
+ if (rxpeer != NULL) {
+ msg->msg_rxpeer = NULL;
+ lnet_peer_decref_locked(rxpeer);
+ }
}
static int
{
lnet_peer_t *p1 = r1->lr_gateway;
lnet_peer_t *p2 = r2->lr_gateway;
+ int r1_hops = (r1->lr_hops == LNET_UNDEFINED_HOPS) ? 1 : r1->lr_hops;
+ int r2_hops = (r2->lr_hops == LNET_UNDEFINED_HOPS) ? 1 : r2->lr_hops;
- if (r1->lr_hops < r2->lr_hops)
+ if (r1->lr_priority < r2->lr_priority)
return 1;
- if (r1->lr_hops > r2->lr_hops)
+ if (r1->lr_priority > r2->lr_priority)
+ return -1;
+
+ if (r1_hops < r2_hops)
+ return 1;
+
+ if (r1_hops > r2_hops)
return -1;
if (p1->lp_txqnob < p2->lp_txqnob)
if (p1->lp_txcredits < p2->lp_txcredits)
return -1;
- return 0;
+ if (r1->lr_seq - r2->lr_seq <= 0)
+ return 1;
+
+ return -1;
}
static lnet_peer_t *
-lnet_find_route_locked(lnet_ni_t *ni, lnet_nid_t target)
+lnet_find_route_locked(lnet_ni_t *ni, lnet_nid_t target, lnet_nid_t rtr_nid)
{
lnet_remotenet_t *rnet;
- lnet_route_t *rtr;
- lnet_route_t *rtr_best;
- lnet_route_t *rtr_last;
+ lnet_route_t *route;
+ lnet_route_t *best_route;
+ lnet_route_t *last_route;
struct lnet_peer *lp_best;
struct lnet_peer *lp;
int rc;
+ /* If @rtr_nid is not LNET_NID_ANY, return the gateway with
+ * rtr_nid nid, otherwise find the best gateway I can use */
+
rnet = lnet_find_net_locked(LNET_NIDNET(target));
if (rnet == NULL)
return NULL;
lp_best = NULL;
- rtr_best = rtr_last = NULL;
- cfs_list_for_each_entry(rtr, &rnet->lrn_routes, lr_list) {
- lp = rtr->lr_gateway;
+ best_route = last_route = NULL;
+ list_for_each_entry(route, &rnet->lrn_routes, lr_list) {
+ lp = route->lr_gateway;
- if (!lp->lp_alive || /* gateway is down */
- (lp->lp_ping_version == LNET_PROTO_PING_VERSION &&
- rtr->lr_downis != 0)) /* NI to target is down */
+ if (!lnet_is_route_alive(route))
continue;
if (ni != NULL && lp->lp_ni != ni)
continue;
+ if (lp->lp_nid == rtr_nid) /* it's pre-determined router */
+ return lp;
+
if (lp_best == NULL) {
- rtr_best = rtr_last = rtr;
+ best_route = last_route = route;
lp_best = lp;
continue;
}
- rc = lnet_compare_routes(rtr, rtr_best);
+ /* no protection on below fields, but it's harmless */
+ if (last_route->lr_seq - route->lr_seq < 0)
+ last_route = route;
+
+ rc = lnet_compare_routes(route, best_route);
if (rc < 0)
continue;
- rtr_best = rtr;
+ best_route = route;
lp_best = lp;
}
- if (rtr_best != NULL) {
- /* Place selected route at the end of the route list to ensure
- * fairness; everything else being equal... */
- cfs_list_del(&rtr_best->lr_list);
- cfs_list_add_tail(&rtr_best->lr_list, &rnet->lrn_routes);
- }
-
+ /* set sequence number on the best router to the latest sequence + 1
+ * so we can round-robin all routers, it's race and inaccurate but
+ * harmless and functional */
+ if (best_route != NULL)
+ best_route->lr_seq = last_route->lr_seq + 1;
return lp_best;
}
int
-lnet_send(lnet_nid_t src_nid, lnet_msg_t *msg)
+lnet_send(lnet_nid_t src_nid, lnet_msg_t *msg, lnet_nid_t rtr_nid)
{
- lnet_nid_t dst_nid = msg->msg_target.nid;
- lnet_ni_t *src_ni;
- lnet_ni_t *local_ni;
- lnet_peer_t *lp;
- int rc;
+ lnet_nid_t dst_nid = msg->msg_target.nid;
+ struct lnet_ni *src_ni;
+ struct lnet_ni *local_ni;
+ struct lnet_peer *lp;
+ int cpt;
+ int cpt2;
+ int rc;
+ /* NB: rtr_nid is set to LNET_NID_ANY for all current use-cases,
+ * but we might want to use pre-determined router for ACK/REPLY
+ * in the future */
+ /* NB: ni != NULL == interface pre-determined (ACK/REPLY) */
LASSERT (msg->msg_txpeer == NULL);
LASSERT (!msg->msg_sending);
LASSERT (!msg->msg_target_is_router);
msg->msg_sending = 1;
- /* NB! ni != NULL == interface pre-determined (ACK/REPLY) */
-
- LNET_LOCK();
+ LASSERT(!msg->msg_tx_committed);
+ cpt = lnet_cpt_of_nid(rtr_nid == LNET_NID_ANY ? dst_nid : rtr_nid);
+ again:
+ lnet_net_lock(cpt);
- if (the_lnet.ln_shutdown) {
- LNET_UNLOCK();
- return -ESHUTDOWN;
- }
+ if (the_lnet.ln_shutdown) {
+ lnet_net_unlock(cpt);
+ return -ESHUTDOWN;
+ }
- if (src_nid == LNET_NID_ANY) {
- src_ni = NULL;
- } else {
- src_ni = lnet_nid2ni_locked(src_nid);
- if (src_ni == NULL) {
- LNET_UNLOCK();
+ if (src_nid == LNET_NID_ANY) {
+ src_ni = NULL;
+ } else {
+ src_ni = lnet_nid2ni_locked(src_nid, cpt);
+ if (src_ni == NULL) {
+ lnet_net_unlock(cpt);
LCONSOLE_WARN("Can't send to %s: src %s is not a "
"local nid\n", libcfs_nid2str(dst_nid),
libcfs_nid2str(src_nid));
LASSERT (!msg->msg_routing);
}
- lnet_msg_commit(msg, 1);
/* Is this for someone on a local network? */
- local_ni = lnet_net2ni_locked(LNET_NIDNET(dst_nid));
+ local_ni = lnet_net2ni_locked(LNET_NIDNET(dst_nid), cpt);
if (local_ni != NULL) {
if (src_ni == NULL) {
src_ni = local_ni;
src_nid = src_ni->ni_nid;
} else if (src_ni == local_ni) {
- lnet_ni_decref_locked(local_ni);
- } else {
- lnet_ni_decref_locked(local_ni);
- lnet_ni_decref_locked(src_ni);
- LNET_UNLOCK();
- LCONSOLE_WARN("No route to %s via from %s\n",
- libcfs_nid2str(dst_nid),
- libcfs_nid2str(src_nid));
- return -EINVAL;
- }
+ lnet_ni_decref_locked(local_ni, cpt);
+ } else {
+ lnet_ni_decref_locked(local_ni, cpt);
+ lnet_ni_decref_locked(src_ni, cpt);
+ lnet_net_unlock(cpt);
+ LCONSOLE_WARN("No route to %s via from %s\n",
+ libcfs_nid2str(dst_nid),
+ libcfs_nid2str(src_nid));
+ return -EINVAL;
+ }
- LASSERT (src_nid != LNET_NID_ANY);
+ LASSERT(src_nid != LNET_NID_ANY);
+ lnet_msg_commit(msg, cpt);
- if (!msg->msg_routing)
- msg->msg_hdr.src_nid = cpu_to_le64(src_nid);
+ if (!msg->msg_routing)
+ msg->msg_hdr.src_nid = cpu_to_le64(src_nid);
- if (src_ni == the_lnet.ln_loni) {
- /* No send credit hassles with LOLND */
- LNET_UNLOCK();
- lnet_ni_send(src_ni, msg);
- lnet_ni_decref(src_ni);
- return 0;
- }
+ if (src_ni == the_lnet.ln_loni) {
+ /* No send credit hassles with LOLND */
+ lnet_net_unlock(cpt);
+ lnet_ni_send(src_ni, msg);
- rc = lnet_nid2peer_locked(&lp, dst_nid);
- lnet_ni_decref_locked(src_ni); /* lp has ref on src_ni; lose mine */
- if (rc != 0) {
- LNET_UNLOCK();
+ lnet_net_lock(cpt);
+ lnet_ni_decref_locked(src_ni, cpt);
+ lnet_net_unlock(cpt);
+ return 0;
+ }
+
+ rc = lnet_nid2peer_locked(&lp, dst_nid, cpt);
+ /* lp has ref on src_ni; lose mine */
+ lnet_ni_decref_locked(src_ni, cpt);
+ if (rc != 0) {
+ lnet_net_unlock(cpt);
LCONSOLE_WARN("Error %d finding peer %s\n", rc,
libcfs_nid2str(dst_nid));
/* ENOMEM or shutting down */
}
LASSERT (lp->lp_ni == src_ni);
} else {
-#ifndef __KERNEL__
- LNET_UNLOCK();
-
- /* NB
- * - once application finishes computation, check here to update
- * router states before it waits for pending IO in LNetEQPoll
- * - recursion breaker: router checker sends no message
- * to remote networks */
- if (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING)
- lnet_router_checker();
-
- LNET_LOCK();
-#endif
- /* sending to a remote network */
- lp = lnet_find_route_locked(src_ni, dst_nid);
- if (lp == NULL) {
- if (src_ni != NULL)
- lnet_ni_decref_locked(src_ni);
- LNET_UNLOCK();
+ /* sending to a remote network */
+ lp = lnet_find_route_locked(src_ni, dst_nid, rtr_nid);
+ if (lp == NULL) {
+ if (src_ni != NULL)
+ lnet_ni_decref_locked(src_ni, cpt);
+ lnet_net_unlock(cpt);
LCONSOLE_WARN("No route to %s via %s "
"(all routers down)\n",
return -EHOSTUNREACH;
}
+ /* rtr_nid is LNET_NID_ANY or NID of pre-determined router,
+ * it's possible that rtr_nid isn't LNET_NID_ANY and lp isn't
+ * pre-determined router, this can happen if router table
+ * was changed when we release the lock */
+ if (rtr_nid != lp->lp_nid) {
+ cpt2 = lnet_cpt_of_nid_locked(lp->lp_nid);
+ if (cpt2 != cpt) {
+ if (src_ni != NULL)
+ lnet_ni_decref_locked(src_ni, cpt);
+ lnet_net_unlock(cpt);
+
+ rtr_nid = lp->lp_nid;
+ cpt = cpt2;
+ goto again;
+ }
+ }
+
CDEBUG(D_NET, "Best route to %s via %s for %s %d\n",
libcfs_nid2str(dst_nid), libcfs_nid2str(lp->lp_nid),
lnet_msgtyp2str(msg->msg_type), msg->msg_len);
src_nid = src_ni->ni_nid;
} else {
LASSERT (src_ni == lp->lp_ni);
- lnet_ni_decref_locked(src_ni);
- }
+ lnet_ni_decref_locked(src_ni, cpt);
+ }
- lnet_peer_addref_locked(lp);
+ lnet_peer_addref_locked(lp);
- LASSERT (src_nid != LNET_NID_ANY);
+ LASSERT(src_nid != LNET_NID_ANY);
+ lnet_msg_commit(msg, cpt);
if (!msg->msg_routing) {
/* I'm the source and now I know which NI to send on */
msg->msg_target_is_router = 1;
msg->msg_target.nid = lp->lp_nid;
- msg->msg_target.pid = LUSTRE_SRV_LNET_PID;
+ msg->msg_target.pid = LNET_PID_LUSTRE;
}
/* 'lp' is our best choice of peer */
msg->msg_txpeer = lp; /* msg takes my ref on lp */
rc = lnet_post_send_locked(msg, 0);
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
- if (rc == EHOSTUNREACH)
- return -EHOSTUNREACH;
+ if (rc < 0)
+ return rc;
- if (rc == 0)
+ if (rc == LNET_CREDIT_OK)
lnet_ni_send(src_ni, msg);
- return 0;
+ return 0; /* rc == LNET_CREDIT_OK or LNET_CREDIT_WAIT */
}
-static void
-lnet_drop_message (lnet_ni_t *ni, void *private, unsigned int nob)
+void
+lnet_drop_message(lnet_ni_t *ni, int cpt, void *private, unsigned int nob)
{
- LNET_LOCK();
- the_lnet.ln_counters->drop_count++;
- the_lnet.ln_counters->drop_length += nob;
- LNET_UNLOCK();
+ lnet_net_lock(cpt);
+ the_lnet.ln_counters[cpt]->drop_count++;
+ the_lnet.ln_counters[cpt]->drop_length += nob;
+ lnet_net_unlock(cpt);
lnet_ni_recv(ni, private, NULL, 0, 0, 0, nob);
}
lnet_build_msg_event(msg, LNET_EVENT_PUT);
- /* Must I ACK? If so I'll grab the ack_wmd out of the header and put
- * it back into the ACK during lnet_finalize() */
+ /* Must I ACK? If so I'll grab the ack_wmd out of the header and put
+ * it back into the ACK during lnet_finalize() */
msg->msg_ack = (!lnet_is_wire_handle_none(&hdr->msg.put.ack_wmd) &&
(msg->msg_md->md_options & LNET_MD_ACK_DISABLE) == 0);
lnet_hdr_t *hdr = &msg->msg_hdr;
struct lnet_match_info info;
int rc;
+ bool ready_delay;
/* Convert put fields to host byte order */
hdr->msg.put.match_bits = le64_to_cpu(hdr->msg.put.match_bits);
info.mi_mbits = hdr->msg.put.match_bits;
msg->msg_rx_ready_delay = ni->ni_lnd->lnd_eager_recv == NULL;
+ ready_delay = msg->msg_rx_ready_delay;
again:
rc = lnet_ptl_match_md(&info, msg);
return 0;
case LNET_MATCHMD_NONE:
- if (msg->msg_rx_delayed) /* attached on delayed list */
+ if (ready_delay)
+ /* no eager_recv or has already called it, should
+ * have been attached on delayed list */
return 0;
rc = lnet_ni_eager_recv(ni, msg);
- if (rc == 0)
+ if (rc == 0) {
+ ready_delay = true;
goto again;
+ }
/* fall through */
case LNET_MATCHMD_DROP:
lnet_ni_recv(ni, msg->msg_private, NULL, 0, 0, 0, 0);
msg->msg_receiving = 0;
- rc = lnet_send(ni->ni_nid, msg);
- if (rc < 0) {
- /* didn't get as far as lnet_ni_send() */
- CERROR("%s: Unable to send REPLY for GET from %s: %d\n",
+ rc = lnet_send(ni->ni_nid, msg, LNET_NID_ANY);
+ if (rc < 0) {
+ /* didn't get as far as lnet_ni_send() */
+ CERROR("%s: Unable to send REPLY for GET from %s: %d\n",
libcfs_nid2str(ni->ni_nid),
libcfs_id2str(info.mi_id), rc);
}
CDEBUG(D_NET, "%s: Reply from %s of length %d/%d into md "LPX64"\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
mlength, rlength, hdr->msg.reply.dst_wmd.wh_object_cookie);
lnet_msg_attach_md(msg, md, 0, mlength);
- if (mlength != 0)
- lnet_setpayloadbuffer(msg);
+ if (mlength != 0)
+ lnet_setpayloadbuffer(msg);
lnet_res_unlock(cpt);
lnet_build_msg_event(msg, LNET_EVENT_REPLY);
- lnet_ni_recv(ni, private, msg, 0, 0, mlength, rlength);
- return 0;
+ lnet_ni_recv(ni, private, msg, 0, 0, mlength, rlength);
+ return 0;
}
static int
}
CDEBUG(D_NET, "%s: ACK from %s into md "LPX64"\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(src),
hdr->msg.ack.dst_wmd.wh_object_cookie);
lnet_msg_attach_md(msg, md, 0, 0);
return 0;
}
-static int
+/**
+ * \retval LNET_CREDIT_OK If \a msg is forwarded
+ * \retval LNET_CREDIT_WAIT If \a msg is blocked because w/o buffer
+ * \retval -ve error code
+ */
+int
lnet_parse_forward_locked(lnet_ni_t *ni, lnet_msg_t *msg)
{
int rc = 0;
-#ifdef __KERNEL__
+ if (!the_lnet.ln_routing)
+ return -ECANCELED;
+
if (msg->msg_rxpeer->lp_rtrcredits <= 0 ||
lnet_msg2bufpool(msg)->rbp_credits <= 0) {
if (ni->ni_lnd->lnd_eager_recv == NULL) {
msg->msg_rx_ready_delay = 1;
} else {
- LNET_UNLOCK();
+ lnet_net_unlock(msg->msg_rx_cpt);
rc = lnet_ni_eager_recv(ni, msg);
- LNET_LOCK();
+ lnet_net_lock(msg->msg_rx_cpt);
}
}
if (rc == 0)
rc = lnet_post_routed_recv_locked(msg, 0);
-#else
- LBUG();
-#endif
+ return rc;
+}
+
+int
+lnet_parse_local(lnet_ni_t *ni, lnet_msg_t *msg)
+{
+ int rc;
+
+ switch (msg->msg_type) {
+ case LNET_MSG_ACK:
+ rc = lnet_parse_ack(ni, msg);
+ break;
+ case LNET_MSG_PUT:
+ rc = lnet_parse_put(ni, msg);
+ break;
+ case LNET_MSG_GET:
+ rc = lnet_parse_get(ni, msg, msg->msg_rdma_get);
+ break;
+ case LNET_MSG_REPLY:
+ rc = lnet_parse_reply(ni, msg);
+ break;
+ default: /* prevent an unused label if !kernel */
+ LASSERT(0);
+ return -EPROTO;
+ }
+
+ LASSERT(rc == 0 || rc == ENOENT);
return rc;
}
}
int
-lnet_parse(lnet_ni_t *ni, lnet_hdr_t *hdr, lnet_nid_t from_nid,
- void *private, int rdma_req)
+lnet_parse(lnet_ni_t *ni, lnet_hdr_t *hdr, lnet_nid_t from_nid,
+ void *private, int rdma_req)
{
- int rc = 0;
- int for_me;
- lnet_msg_t *msg;
- lnet_pid_t dest_pid;
- lnet_nid_t dest_nid;
- lnet_nid_t src_nid;
- __u32 payload_length;
- __u32 type;
-
- LASSERT (!cfs_in_interrupt ());
-
- type = le32_to_cpu(hdr->type);
- src_nid = le64_to_cpu(hdr->src_nid);
- dest_nid = le64_to_cpu(hdr->dest_nid);
- dest_pid = le32_to_cpu(hdr->dest_pid);
- payload_length = le32_to_cpu(hdr->payload_length);
-
- for_me = (ni->ni_nid == dest_nid);
-
- switch (type) {
- case LNET_MSG_ACK:
- case LNET_MSG_GET:
- if (payload_length > 0) {
- CERROR("%s, src %s: bad %s payload %d (0 expected)\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid),
- lnet_msgtyp2str(type), payload_length);
- return -EPROTO;
- }
- break;
-
- case LNET_MSG_PUT:
- case LNET_MSG_REPLY:
- if (payload_length > (__u32)(for_me ? LNET_MAX_PAYLOAD : LNET_MTU)) {
- CERROR("%s, src %s: bad %s payload %d "
- "(%d max expected)\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid),
- lnet_msgtyp2str(type),
- payload_length,
- for_me ? LNET_MAX_PAYLOAD : LNET_MTU);
- return -EPROTO;
- }
- break;
+ int rc = 0;
+ int cpt;
+ int for_me;
+ struct lnet_msg *msg;
+ lnet_pid_t dest_pid;
+ lnet_nid_t dest_nid;
+ lnet_nid_t src_nid;
+ __u32 payload_length;
+ __u32 type;
+
+ LASSERT (!in_interrupt ());
+
+ type = le32_to_cpu(hdr->type);
+ src_nid = le64_to_cpu(hdr->src_nid);
+ dest_nid = le64_to_cpu(hdr->dest_nid);
+ dest_pid = le32_to_cpu(hdr->dest_pid);
+ payload_length = le32_to_cpu(hdr->payload_length);
+
+ for_me = (ni->ni_nid == dest_nid);
+ cpt = lnet_cpt_of_nid(from_nid);
+
+ switch (type) {
+ case LNET_MSG_ACK:
+ case LNET_MSG_GET:
+ if (payload_length > 0) {
+ CERROR("%s, src %s: bad %s payload %d (0 expected)\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid),
+ lnet_msgtyp2str(type), payload_length);
+ return -EPROTO;
+ }
+ break;
+
+ case LNET_MSG_PUT:
+ case LNET_MSG_REPLY:
+ if (payload_length >
+ (__u32)(for_me ? LNET_MAX_PAYLOAD : LNET_MTU)) {
+ CERROR("%s, src %s: bad %s payload %d "
+ "(%d max expected)\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid),
+ lnet_msgtyp2str(type),
+ payload_length,
+ for_me ? LNET_MAX_PAYLOAD : LNET_MTU);
+ return -EPROTO;
+ }
+ break;
- default:
- CERROR("%s, src %s: Bad message type 0x%x\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid), type);
- return -EPROTO;
- }
+ default:
+ CERROR("%s, src %s: Bad message type 0x%x\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid), type);
+ return -EPROTO;
+ }
if (the_lnet.ln_routing &&
ni->ni_last_alive != cfs_time_current_sec()) {
- LNET_LOCK();
-
/* NB: so far here is the only place to set NI status to "up */
+ lnet_ni_lock(ni);
ni->ni_last_alive = cfs_time_current_sec();
if (ni->ni_status != NULL &&
ni->ni_status->ns_status == LNET_NI_STATUS_DOWN)
ni->ni_status->ns_status = LNET_NI_STATUS_UP;
- LNET_UNLOCK();
- }
+ lnet_ni_unlock(ni);
+ }
- /* Regard a bad destination NID as a protocol error. Senders should
- * know what they're doing; if they don't they're misconfigured, buggy
- * or malicious so we chop them off at the knees :) */
-
- if (!for_me) {
- if (LNET_NIDNET(dest_nid) == LNET_NIDNET(ni->ni_nid)) {
- /* should have gone direct */
- CERROR ("%s, src %s: Bad dest nid %s "
- "(should have been sent direct)\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid),
- libcfs_nid2str(dest_nid));
- return -EPROTO;
- }
+ /* Regard a bad destination NID as a protocol error. Senders should
+ * know what they're doing; if they don't they're misconfigured, buggy
+ * or malicious so we chop them off at the knees :) */
- if (lnet_islocalnid(dest_nid)) {
- /* dest is another local NI; sender should have used
- * this node's NID on its own network */
- CERROR ("%s, src %s: Bad dest nid %s "
- "(it's my nid but on a different network)\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid),
- libcfs_nid2str(dest_nid));
- return -EPROTO;
- }
+ if (!for_me) {
+ if (LNET_NIDNET(dest_nid) == LNET_NIDNET(ni->ni_nid)) {
+ /* should have gone direct */
+ CERROR("%s, src %s: Bad dest nid %s "
+ "(should have been sent direct)\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid),
+ libcfs_nid2str(dest_nid));
+ return -EPROTO;
+ }
- if (rdma_req && type == LNET_MSG_GET) {
- CERROR ("%s, src %s: Bad optimized GET for %s "
- "(final destination must be me)\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid),
- libcfs_nid2str(dest_nid));
- return -EPROTO;
- }
+ if (lnet_islocalnid(dest_nid)) {
+ /* dest is another local NI; sender should have used
+ * this node's NID on its own network */
+ CERROR("%s, src %s: Bad dest nid %s "
+ "(it's my nid but on a different network)\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid),
+ libcfs_nid2str(dest_nid));
+ return -EPROTO;
+ }
- if (!the_lnet.ln_routing) {
- CERROR ("%s, src %s: Dropping message for %s "
- "(routing not enabled)\n",
- libcfs_nid2str(from_nid),
- libcfs_nid2str(src_nid),
- libcfs_nid2str(dest_nid));
- goto drop;
- }
- }
+ if (rdma_req && type == LNET_MSG_GET) {
+ CERROR("%s, src %s: Bad optimized GET for %s "
+ "(final destination must be me)\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid),
+ libcfs_nid2str(dest_nid));
+ return -EPROTO;
+ }
- /* Message looks OK; we're not going to return an error, so we MUST
- * call back lnd_recv() come what may... */
+ if (!the_lnet.ln_routing) {
+ CERROR("%s, src %s: Dropping message for %s "
+ "(routing not enabled)\n",
+ libcfs_nid2str(from_nid),
+ libcfs_nid2str(src_nid),
+ libcfs_nid2str(dest_nid));
+ goto drop;
+ }
+ }
- if (!cfs_list_empty (&the_lnet.ln_test_peers) && /* normally we don't */
- fail_peer (src_nid, 0)) /* shall we now? */
- {
- CERROR("%s, src %s: Dropping %s to simulate failure\n",
- libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
- lnet_msgtyp2str(type));
- goto drop;
- }
+ /* Message looks OK; we're not going to return an error, so we MUST
+ * call back lnd_recv() come what may... */
- msg = lnet_msg_alloc();
- if (msg == NULL) {
- CERROR("%s, src %s: Dropping %s (out of memory)\n",
- libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
- lnet_msgtyp2str(type));
- goto drop;
- }
+ if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
+ fail_peer(src_nid, 0)) { /* shall we now? */
+ CERROR("%s, src %s: Dropping %s to simulate failure\n",
+ libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
+ lnet_msgtyp2str(type));
+ goto drop;
+ }
- /* msg zeroed in lnet_msg_alloc; i.e. flags all clear, pointers NULL etc */
+ if (!list_empty(&the_lnet.ln_drop_rules) &&
+ lnet_drop_rule_match(hdr)) {
+ CDEBUG(D_NET, "%s, src %s, dst %s: Dropping %s to simulate"
+ "silent message loss\n",
+ libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
+ libcfs_nid2str(dest_nid), lnet_msgtyp2str(type));
+ goto drop;
+ }
- msg->msg_type = type;
- msg->msg_private = private;
- msg->msg_receiving = 1;
- msg->msg_len = msg->msg_wanted = payload_length;
- msg->msg_offset = 0;
- msg->msg_hdr = *hdr;
+
+ msg = lnet_msg_alloc();
+ if (msg == NULL) {
+ CERROR("%s, src %s: Dropping %s (out of memory)\n",
+ libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
+ lnet_msgtyp2str(type));
+ goto drop;
+ }
+
+ /* msg zeroed in lnet_msg_alloc; i.e. flags all clear,
+ * pointers NULL etc */
+
+ msg->msg_type = type;
+ msg->msg_private = private;
+ msg->msg_receiving = 1;
+ msg->msg_rdma_get = rdma_req;
+ msg->msg_len = msg->msg_wanted = payload_length;
+ msg->msg_offset = 0;
+ msg->msg_hdr = *hdr;
/* for building message event */
msg->msg_from = from_nid;
if (!for_me) {
msg->msg_hdr.payload_length = payload_length;
}
- LNET_LOCK();
- rc = lnet_nid2peer_locked(&msg->msg_rxpeer, from_nid);
- if (rc != 0) {
- LNET_UNLOCK();
- CERROR("%s, src %s: Dropping %s "
- "(error %d looking up sender)\n",
- libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
- lnet_msgtyp2str(type), rc);
+ lnet_net_lock(cpt);
+ rc = lnet_nid2peer_locked(&msg->msg_rxpeer, from_nid, cpt);
+ if (rc != 0) {
+ lnet_net_unlock(cpt);
+ CERROR("%s, src %s: Dropping %s "
+ "(error %d looking up sender)\n",
+ libcfs_nid2str(from_nid), libcfs_nid2str(src_nid),
+ lnet_msgtyp2str(type), rc);
lnet_msg_free(msg);
goto drop;
}
- lnet_msg_commit(msg, 0);
+ if (lnet_isrouter(msg->msg_rxpeer)) {
+ lnet_peer_set_alive(msg->msg_rxpeer);
+ if (avoid_asym_router_failure &&
+ LNET_NIDNET(src_nid) != LNET_NIDNET(from_nid)) {
+ /* received a remote message from router, update
+ * remote NI status on this router.
+ * NB: multi-hop routed message will be ignored.
+ */
+ lnet_router_ni_update_locked(msg->msg_rxpeer,
+ LNET_NIDNET(src_nid));
+ }
+ }
+
+ lnet_msg_commit(msg, cpt);
+
+ /* message delay simulation */
+ if (unlikely(!list_empty(&the_lnet.ln_delay_rules) &&
+ lnet_delay_rule_match_locked(hdr, msg))) {
+ lnet_net_unlock(cpt);
+ return 0;
+ }
if (!for_me) {
rc = lnet_parse_forward_locked(ni, msg);
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
if (rc < 0)
goto free_drop;
- if (rc == 0) {
+
+ if (rc == LNET_CREDIT_OK) {
lnet_ni_recv(ni, msg->msg_private, msg, 0,
0, payload_length, payload_length);
}
return 0;
}
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
- switch (type) {
- case LNET_MSG_ACK:
- rc = lnet_parse_ack(ni, msg);
- break;
- case LNET_MSG_PUT:
- rc = lnet_parse_put(ni, msg);
- break;
- case LNET_MSG_GET:
- rc = lnet_parse_get(ni, msg, rdma_req);
- break;
- case LNET_MSG_REPLY:
- rc = lnet_parse_reply(ni, msg);
- break;
- default:
- LASSERT(0);
- rc = -EPROTO;
- goto free_drop; /* prevent an unused label if !kernel */
- }
-
- if (rc == 0)
- return 0;
-
- LASSERT (rc == ENOENT);
+ rc = lnet_parse_local(ni, msg);
+ if (rc != 0)
+ goto free_drop;
+ return 0;
free_drop:
LASSERT(msg->msg_md == NULL);
lnet_finalize(ni, msg, rc);
drop:
- lnet_drop_message(ni, private, payload_length);
+ lnet_drop_message(ni, cpt, private, payload_length);
return 0;
}
+EXPORT_SYMBOL(lnet_parse);
void
-lnet_drop_delayed_msg_list(cfs_list_t *head, char *reason)
+lnet_drop_delayed_msg_list(struct list_head *head, char *reason)
{
- while (!cfs_list_empty(head)) {
+ while (!list_empty(head)) {
lnet_process_id_t id = {0};
lnet_msg_t *msg;
- msg = cfs_list_entry(head->next, lnet_msg_t, msg_list);
- cfs_list_del(&msg->msg_list);
+ msg = list_entry(head->next, lnet_msg_t, msg_list);
+ list_del(&msg->msg_list);
id.nid = msg->msg_hdr.src_nid;
id.pid = msg->msg_hdr.src_pid;
* until that's done */
lnet_drop_message(msg->msg_rxpeer->lp_ni,
+ msg->msg_rxpeer->lp_cpt,
msg->msg_private, msg->msg_len);
-
- LNET_LOCK();
- lnet_peer_decref_locked(msg->msg_rxpeer);
- LNET_UNLOCK();
-
- lnet_msg_free(msg);
+ /*
+ * NB: message will not generate event because w/o attached MD,
+ * but we still should give error code so lnet_msg_decommit()
+ * can skip counters operations and other checks.
+ */
+ lnet_finalize(msg->msg_rxpeer->lp_ni, msg, -ENOENT);
}
}
void
-lnet_recv_delayed_msg_list(cfs_list_t *head)
+lnet_recv_delayed_msg_list(struct list_head *head)
{
- while (!cfs_list_empty(head)) {
+ while (!list_empty(head)) {
lnet_msg_t *msg;
lnet_process_id_t id;
- msg = cfs_list_entry(head->next, lnet_msg_t, msg_list);
- cfs_list_del(&msg->msg_list);
+ msg = list_entry(head->next, lnet_msg_t, msg_list);
+ list_del(&msg->msg_list);
/* md won't disappear under me, since each msg
* holds a ref on it */
int cpt;
int rc;
- LASSERT (the_lnet.ln_init);
- LASSERT (the_lnet.ln_refcount > 0);
+ LASSERT(the_lnet.ln_refcount > 0);
- if (!cfs_list_empty (&the_lnet.ln_test_peers) && /* normally we don't */
- fail_peer (target.nid, 1)) /* shall we now? */
- {
+ if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
+ fail_peer(target.nid, 1)) { /* shall we now? */
CERROR("Dropping PUT to %s: simulated failure\n",
libcfs_id2str(target));
return -EIO;
libcfs_id2str(target));
return -ENOMEM;
}
- msg->msg_vmflush = !!cfs_memory_pressure_get();
+ msg->msg_vmflush = !!memory_pressure_get();
cpt = lnet_cpt_of_cookie(mdh.cookie);
lnet_res_lock(cpt);
lnet_res_unlock(cpt);
lnet_msg_free(msg);
- return -ENOENT;
- }
+ return -ENOENT;
+ }
- CDEBUG(D_NET, "LNetPut -> %s\n", libcfs_id2str(target));
+ CDEBUG(D_NET, "LNetPut -> %s\n", libcfs_id2str(target));
lnet_msg_attach_md(msg, md, 0, 0);
/* NB handles only looked up by creator (no flips) */
if (ack == LNET_ACK_REQ) {
- msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
the_lnet.ln_interface_cookie;
- msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
md->md_lh.lh_cookie;
} else {
- msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_interface_cookie =
LNET_WIRE_HANDLE_COOKIE_NONE;
- msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
+ msg->msg_hdr.msg.put.ack_wmd.wh_object_cookie =
LNET_WIRE_HANDLE_COOKIE_NONE;
}
lnet_build_msg_event(msg, LNET_EVENT_SEND);
- rc = lnet_send(self, msg);
+ rc = lnet_send(self, msg, LNET_NID_ANY);
if (rc != 0) {
CNETERR( "Error sending PUT to %s: %d\n",
libcfs_id2str(target), rc);
/* completion will be signalled by an event */
return 0;
}
+EXPORT_SYMBOL(LNetPut);
lnet_msg_t *
lnet_create_reply_msg (lnet_ni_t *ni, lnet_msg_t *getmsg)
LASSERT(!getmsg->msg_target_is_router);
LASSERT(!getmsg->msg_routing);
+ if (msg == NULL) {
+ CERROR("%s: Dropping REPLY from %s: can't allocate msg\n",
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id));
+ goto drop;
+ }
+
cpt = lnet_cpt_of_cookie(getmd->md_lh.lh_cookie);
lnet_res_lock(cpt);
- LASSERT (getmd->md_refcount > 0);
-
- if (msg == NULL) {
- CERROR ("%s: Dropping REPLY from %s: can't allocate msg\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id));
- goto drop;
- }
+ LASSERT(getmd->md_refcount > 0);
if (getmd->md_threshold == 0) {
CERROR ("%s: Dropping REPLY from %s for inactive MD %p\n",
- libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id),
+ libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id),
getmd);
lnet_res_unlock(cpt);
goto drop;
- }
+ }
- LASSERT (getmd->md_offset == 0);
+ LASSERT(getmd->md_offset == 0);
CDEBUG(D_NET, "%s: Reply from %s md %p\n",
libcfs_nid2str(ni->ni_nid), libcfs_id2str(peer_id), getmd);
/* setup information for lnet_build_msg_event */
msg->msg_from = peer_id.nid;
- msg->msg_type = LNET_MSG_GET; /* flag this msg as an "optimized" GET */
+ msg->msg_type = LNET_MSG_GET; /* flag this msg as an "optimized" GET */
msg->msg_hdr.src_nid = peer_id.nid;
msg->msg_hdr.payload_length = getmd->md_length;
msg->msg_receiving = 1; /* required by lnet_msg_attach_md */
lnet_msg_attach_md(msg, getmd, getmd->md_offset, getmd->md_length);
lnet_res_unlock(cpt);
- LNET_LOCK();
- lnet_msg_commit(msg, 0);
- LNET_UNLOCK();
+ cpt = lnet_cpt_of_nid(peer_id.nid);
+
+ lnet_net_lock(cpt);
+ lnet_msg_commit(msg, cpt);
+ lnet_net_unlock(cpt);
lnet_build_msg_event(msg, LNET_EVENT_REPLY);
return msg;
drop:
- LNET_LOCK();
- the_lnet.ln_counters->drop_count++;
- the_lnet.ln_counters->drop_length += getmd->md_length;
- LNET_UNLOCK ();
+ cpt = lnet_cpt_of_nid(peer_id.nid);
+
+ lnet_net_lock(cpt);
+ the_lnet.ln_counters[cpt]->drop_count++;
+ the_lnet.ln_counters[cpt]->drop_length += getmd->md_length;
+ lnet_net_unlock(cpt);
if (msg != NULL)
lnet_msg_free(msg);
return NULL;
}
+EXPORT_SYMBOL(lnet_create_reply_msg);
void
lnet_set_reply_msg_len(lnet_ni_t *ni, lnet_msg_t *reply, unsigned int len)
reply->msg_ev.mlength = len;
}
+EXPORT_SYMBOL(lnet_set_reply_msg_len);
/**
* Initiate an asynchronous GET operation.
* \retval -ENOENT Invalid MD object.
*/
int
-LNetGet(lnet_nid_t self, lnet_handle_md_t mdh,
- lnet_process_id_t target, unsigned int portal,
- __u64 match_bits, unsigned int offset)
+LNetGet(lnet_nid_t self, lnet_handle_md_t mdh,
+ lnet_process_id_t target, unsigned int portal,
+ __u64 match_bits, unsigned int offset)
{
struct lnet_msg *msg;
struct lnet_libmd *md;
int cpt;
int rc;
- LASSERT (the_lnet.ln_init);
LASSERT (the_lnet.ln_refcount > 0);
- if (!cfs_list_empty (&the_lnet.ln_test_peers) && /* normally we don't */
- fail_peer (target.nid, 1)) /* shall we now? */
+ if (!list_empty(&the_lnet.ln_test_peers) && /* normally we don't */
+ fail_peer(target.nid, 1)) /* shall we now? */
{
CERROR("Dropping GET to %s: simulated failure\n",
libcfs_id2str(target));
lnet_res_unlock(cpt);
lnet_msg_free(msg);
-
return -ENOENT;
}
lnet_build_msg_event(msg, LNET_EVENT_SEND);
- rc = lnet_send(self, msg);
- if (rc < 0) {
- CNETERR( "Error sending GET to %s: %d\n",
- libcfs_id2str(target), rc);
- lnet_finalize (NULL, msg, rc);
- }
+ rc = lnet_send(self, msg, LNET_NID_ANY);
+ if (rc < 0) {
+ CNETERR("Error sending GET to %s: %d\n",
+ libcfs_id2str(target), rc);
+ lnet_finalize(NULL, msg, rc);
+ }
/* completion will be signalled by an event */
return 0;
}
+EXPORT_SYMBOL(LNetGet);
/**
* Calculate distance to node at \a dstnid.
* \retval -EHOSTUNREACH If \a dstnid is not reachable.
*/
int
-LNetDist (lnet_nid_t dstnid, lnet_nid_t *srcnidp, __u32 *orderp)
+LNetDist(lnet_nid_t dstnid, lnet_nid_t *srcnidp, __u32 *orderp)
{
- cfs_list_t *e;
- lnet_ni_t *ni;
- lnet_remotenet_t *rnet;
- __u32 dstnet = LNET_NIDNET(dstnid);
- int hops;
- __u32 order = 2;
+ struct list_head *e;
+ struct lnet_ni *ni;
+ lnet_remotenet_t *rnet;
+ __u32 dstnet = LNET_NIDNET(dstnid);
+ int hops;
+ int cpt;
+ __u32 order = 2;
+ struct list_head *rn_list;
/* if !local_nid_dist_zero, I don't return a distance of 0 ever
* (when lustre sees a distance of 0, it substitutes 0@lo), so I
* keep order 0 free for 0@lo and order 1 free for a local NID
* match */
- LASSERT (the_lnet.ln_init);
LASSERT (the_lnet.ln_refcount > 0);
- LNET_LOCK();
+ cpt = lnet_net_lock_current();
- cfs_list_for_each (e, &the_lnet.ln_nis) {
- ni = cfs_list_entry(e, lnet_ni_t, ni_list);
+ list_for_each(e, &the_lnet.ln_nis) {
+ ni = list_entry(e, lnet_ni_t, ni_list);
if (ni->ni_nid == dstnid) {
if (srcnidp != NULL)
else
*orderp = 1;
}
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
return local_nid_dist_zero ? 0 : 1;
}
*srcnidp = ni->ni_nid;
if (orderp != NULL)
*orderp = order;
- LNET_UNLOCK();
- return 1;
- }
-
- order++;
- }
-
- cfs_list_for_each (e, &the_lnet.ln_remote_nets) {
- rnet = cfs_list_entry(e, lnet_remotenet_t, lrn_list);
-
- if (rnet->lrn_net == dstnet) {
- lnet_route_t *route;
- lnet_route_t *shortest = NULL;
-
- LASSERT (!cfs_list_empty(&rnet->lrn_routes));
-
- cfs_list_for_each_entry(route, &rnet->lrn_routes,
- lr_list) {
- if (shortest == NULL ||
- route->lr_hops < shortest->lr_hops)
- shortest = route;
- }
-
- LASSERT (shortest != NULL);
- hops = shortest->lr_hops;
- if (srcnidp != NULL)
- *srcnidp = shortest->lr_gateway->lp_ni->ni_nid;
- if (orderp != NULL)
- *orderp = order;
- LNET_UNLOCK();
- return hops + 1;
- }
- order++;
- }
-
- LNET_UNLOCK();
- return -EHOSTUNREACH;
-}
-
-/**
- * Set the number of asynchronous messages expected from a target process.
- *
- * This function is only meaningful for userspace callers. It's a no-op when
- * called from kernel.
- *
- * Asynchronous messages are those that can come from a target when the
- * userspace process is not waiting for IO to complete; e.g., AST callbacks
- * from Lustre servers. Specifying the expected number of such messages
- * allows them to be eagerly received when user process is not running in
- * LNet; otherwise network errors may occur.
- *
- * \param id Process ID of the target process.
- * \param nasync Number of asynchronous messages expected from the target.
- *
- * \return 0 on success, and an error code otherwise.
- */
-int
-LNetSetAsync(lnet_process_id_t id, int nasync)
-{
-#ifdef __KERNEL__
- return 0;
-#else
- lnet_ni_t *ni;
- lnet_remotenet_t *rnet;
- cfs_list_t *tmp;
- lnet_route_t *route;
- lnet_nid_t *nids;
- int nnids;
- int maxnids = 256;
- int rc = 0;
- int rc2;
-
- /* Target on a local network? */
- ni = lnet_net2ni(LNET_NIDNET(id.nid));
- if (ni != NULL) {
- if (ni->ni_lnd->lnd_setasync != NULL)
- rc = (ni->ni_lnd->lnd_setasync)(ni, id, nasync);
- lnet_ni_decref(ni);
- return rc;
- }
-
- /* Target on a remote network: apply to routers */
- again:
- LIBCFS_ALLOC(nids, maxnids * sizeof(*nids));
- if (nids == NULL)
- return -ENOMEM;
- nnids = 0;
-
- /* Snapshot all the router NIDs */
- LNET_LOCK();
- rnet = lnet_find_net_locked(LNET_NIDNET(id.nid));
- if (rnet != NULL) {
- cfs_list_for_each(tmp, &rnet->lrn_routes) {
- if (nnids == maxnids) {
- LNET_UNLOCK();
- LIBCFS_FREE(nids, maxnids * sizeof(*nids));
- maxnids *= 2;
- goto again;
- }
-
- route = cfs_list_entry(tmp, lnet_route_t, lr_list);
- nids[nnids++] = route->lr_gateway->lp_nid;
- }
- }
- LNET_UNLOCK();
-
- /* set async on all the routers */
- while (nnids-- > 0) {
- id.pid = LUSTRE_SRV_LNET_PID;
- id.nid = nids[nnids];
+ lnet_net_unlock(cpt);
+ return 1;
+ }
- ni = lnet_net2ni(LNET_NIDNET(id.nid));
- if (ni == NULL)
- continue;
+ order++;
+ }
- if (ni->ni_lnd->lnd_setasync != NULL) {
- rc2 = (ni->ni_lnd->lnd_setasync)(ni, id, nasync);
- if (rc2 != 0)
- rc = rc2;
- }
- lnet_ni_decref(ni);
- }
+ rn_list = lnet_net2rnethash(dstnet);
+ list_for_each(e, rn_list) {
+ rnet = list_entry(e, lnet_remotenet_t, lrn_list);
+
+ if (rnet->lrn_net == dstnet) {
+ lnet_route_t *route;
+ lnet_route_t *shortest = NULL;
+ __u32 shortest_hops = LNET_UNDEFINED_HOPS;
+ __u32 route_hops;
+
+ LASSERT(!list_empty(&rnet->lrn_routes));
+
+ list_for_each_entry(route, &rnet->lrn_routes,
+ lr_list) {
+ route_hops = route->lr_hops;
+ if (route_hops == LNET_UNDEFINED_HOPS)
+ route_hops = 1;
+ if (shortest == NULL ||
+ route_hops < shortest_hops) {
+ shortest = route;
+ shortest_hops = route_hops;
+ }
+ }
+
+ LASSERT(shortest != NULL);
+ hops = shortest_hops;
+ if (srcnidp != NULL)
+ *srcnidp = shortest->lr_gateway->lp_ni->ni_nid;
+ if (orderp != NULL)
+ *orderp = order;
+ lnet_net_unlock(cpt);
+ return hops + 1;
+ }
+ order++;
+ }
- LIBCFS_FREE(nids, maxnids * sizeof(*nids));
- return rc;
-#endif
+ lnet_net_unlock(cpt);
+ return -EHOSTUNREACH;
}
+EXPORT_SYMBOL(LNetDist);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff