*/
#include "socknal.h"
-#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
-# include <linux/syscalls.h>
-#endif
/*
* LIB functions follow
PORTAL_FREE(ltx, ltx->ltx_desc_size);
}
-#if (SOCKNAL_ZC && SOCKNAL_VADDR_ZC)
-struct page *
-ksocknal_kvaddr_to_page (unsigned long vaddr)
-{
- struct page *page;
-
- if (vaddr >= VMALLOC_START &&
- vaddr < VMALLOC_END)
- page = vmalloc_to_page ((void *)vaddr);
-#if CONFIG_HIGHMEM
- else if (vaddr >= PKMAP_BASE &&
- vaddr < (PKMAP_BASE + LAST_PKMAP * PAGE_SIZE))
- page = vmalloc_to_page ((void *)vaddr);
- /* in 2.4 ^ just walks the page tables */
-#endif
- else
- page = virt_to_page (vaddr);
-
- if (page == NULL ||
- !VALID_PAGE (page))
- return (NULL);
-
- return (page);
-}
-#endif
-
int
ksocknal_send_iov (ksock_conn_t *conn, ksock_tx_t *tx)
-{
- struct socket *sock = conn->ksnc_sock;
+{
struct iovec *iov = tx->tx_iov;
- int fragsize = iov->iov_len;
- unsigned long vaddr = (unsigned long)iov->iov_base;
- int more = (tx->tx_niov > 1) ||
- (tx->tx_nkiov > 0) ||
- (!list_empty (&conn->ksnc_tx_queue));
-#if (SOCKNAL_ZC && SOCKNAL_VADDR_ZC)
- int offset = vaddr & (PAGE_SIZE - 1);
- int zcsize = MIN (fragsize, PAGE_SIZE - offset);
- struct page *page;
-#endif
- int rc;
+ int nob;
+ int rc;
- /* NB we can't trust socket ops to either consume our iovs
- * or leave them alone, so we only send 1 frag at a time. */
- LASSERT (fragsize <= tx->tx_resid);
LASSERT (tx->tx_niov > 0);
-
-#if (SOCKNAL_ZC && SOCKNAL_VADDR_ZC)
- if (zcsize >= ksocknal_data.ksnd_zc_min_frag &&
- (sock->sk->route_caps & NETIF_F_SG) &&
- (sock->sk->route_caps & (NETIF_F_IP_CSUM | NETIF_F_NO_CSUM | NETIF_F_HW_CSUM)) &&
- (page = ksocknal_kvaddr_to_page (vaddr)) != NULL) {
-
- CDEBUG(D_NET, "vaddr %p, page %p->%p + offset %x for %d\n",
- (void *)vaddr, page, page_address(page), offset, zcsize);
- if (fragsize > zcsize) {
- more = 1;
- fragsize = zcsize;
- }
+ /* Never touch tx->tx_iov inside ksocknal_lib_send_iov() */
+ rc = ksocknal_lib_send_iov(conn, tx);
- rc = tcp_sendpage_zccd(sock, page, offset, zcsize,
- more ? (MSG_DONTWAIT | MSG_MORE) : MSG_DONTWAIT,
- &tx->tx_zccd);
- } else
-#endif
- {
- /* NB don't pass tx's iov; sendmsg may or may not update it */
- struct iovec fragiov = { .iov_base = (void *)vaddr,
- .iov_len = fragsize};
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &fragiov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = more ? (MSG_DONTWAIT | MSG_MORE) : MSG_DONTWAIT
- };
- mm_segment_t oldmm = get_fs();
-
- set_fs (KERNEL_DS);
- rc = sock_sendmsg(sock, &msg, fragsize);
- set_fs (oldmm);
- }
+ if (rc <= 0) /* sent nothing? */
+ return (rc);
- if (rc > 0) {
- tx->tx_resid -= rc;
+ nob = rc;
+ LASSERT (nob <= tx->tx_resid);
+ tx->tx_resid -= nob;
+
+ /* "consume" iov */
+ do {
+ LASSERT (tx->tx_niov > 0);
+
+ if (nob < iov->iov_len) {
+ iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
+ iov->iov_len -= nob;
+ return (rc);
+ }
+
+ nob -= iov->iov_len;
+ tx->tx_iov = ++iov;
+ tx->tx_niov--;
+ } while (nob != 0);
- if (rc < iov->iov_len) {
- /* didn't send whole iov entry... */
- iov->iov_base = (void *)(vaddr + rc);
- iov->iov_len -= rc;
- } else {
- tx->tx_iov++;
- tx->tx_niov--;
- }
- }
-
return (rc);
}
int
ksocknal_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx)
-{
- struct socket *sock = conn->ksnc_sock;
+{
ptl_kiov_t *kiov = tx->tx_kiov;
- int fragsize = kiov->kiov_len;
- struct page *page = kiov->kiov_page;
- int offset = kiov->kiov_offset;
- int more = (tx->tx_nkiov > 1) ||
- (!list_empty (&conn->ksnc_tx_queue));
- int rc;
+ int nob;
+ int rc;
- /* NB we can't trust socket ops to either consume our iovs
- * or leave them alone, so we only send 1 frag at a time. */
- LASSERT (fragsize <= tx->tx_resid);
- LASSERT (offset + fragsize <= PAGE_SIZE);
- LASSERT (tx->tx_niov == 0);
+ LASSERT (tx->tx_niov == 0);
LASSERT (tx->tx_nkiov > 0);
-#if SOCKNAL_ZC
- if (fragsize >= ksocknal_tunables.ksnd_zc_min_frag &&
- (sock->sk->route_caps & NETIF_F_SG) &&
- (sock->sk->route_caps & (NETIF_F_IP_CSUM | NETIF_F_NO_CSUM | NETIF_F_HW_CSUM))) {
+ /* Never touch tx->tx_kiov inside ksocknal_lib_send_kiov() */
+ rc = ksocknal_lib_send_kiov(conn, tx);
- CDEBUG(D_NET, "page %p + offset %x for %d\n",
- page, offset, fragsize);
+ if (rc <= 0) /* sent nothing? */
+ return (rc);
+
+ nob = rc;
+ LASSERT (nob <= tx->tx_resid);
+ tx->tx_resid -= nob;
- rc = tcp_sendpage_zccd(sock, page, offset, fragsize,
- more ? (MSG_DONTWAIT | MSG_MORE) : MSG_DONTWAIT,
- &tx->tx_zccd);
- } else
-#endif
- {
- char *addr = ((char *)kmap (page)) + offset;
- struct iovec fragiov = {.iov_base = addr,
- .iov_len = fragsize};
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &fragiov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = more ? (MSG_DONTWAIT | MSG_MORE) : MSG_DONTWAIT
- };
- mm_segment_t oldmm = get_fs();
-
- set_fs (KERNEL_DS);
- rc = sock_sendmsg(sock, &msg, fragsize);
- set_fs (oldmm);
+ /* "consume" kiov */
+ do {
+ LASSERT(tx->tx_nkiov > 0);
- kunmap (page);
- }
+ if (nob < kiov->kiov_len) {
+ kiov->kiov_offset += nob;
+ kiov->kiov_len -= nob;
+ return rc;
+ }
- if (rc > 0) {
- tx->tx_resid -= rc;
-
- if (rc < fragsize) {
- kiov->kiov_offset = offset + rc;
- kiov->kiov_len = fragsize - rc;
- } else {
- tx->tx_kiov++;
- tx->tx_nkiov--;
- }
- }
+ nob -= kiov->kiov_len;
+ tx->tx_kiov = ++kiov;
+ tx->tx_nkiov--;
+ } while (nob != 0);
return (rc);
}
if (ksocknal_data.ksnd_stall_tx != 0) {
set_current_state (TASK_UNINTERRUPTIBLE);
- schedule_timeout (ksocknal_data.ksnd_stall_tx * HZ);
+ schedule_timeout (cfs_time_seconds(ksocknal_data.ksnd_stall_tx));
}
LASSERT (tx->tx_resid != 0);
rc = ksocknal_send_kiov (conn, tx);
}
- bufnob = conn->ksnc_sock->sk->sk_wmem_queued;
+ bufnob = SOCK_WMEM_QUEUED(conn->ksnc_sock);
if (rc > 0) /* sent something? */
conn->ksnc_tx_bufnob += rc; /* account it */
if (bufnob < conn->ksnc_tx_bufnob) {
/* allocated send buffer bytes < computed; infer
* something got ACKed */
- conn->ksnc_tx_deadline = jiffies +
- ksocknal_tunables.ksnd_io_timeout * HZ;
- conn->ksnc_peer->ksnp_last_alive = jiffies;
+ conn->ksnc_tx_deadline = cfs_time_shift(ksocknal_tunables.ksnd_io_timeout);
+ conn->ksnc_peer->ksnp_last_alive = cfs_time_current();
conn->ksnc_tx_bufnob = bufnob;
mb();
}
- if (rc <= 0) {
- /* Didn't write anything.
- *
- * NB: rc == 0 and rc == -EAGAIN both mean try
- * again later (linux stack returns -EAGAIN for
- * this, but Adaptech TOE returns 0).
- *
- * Also, sends never fail with -ENOMEM, just
- * -EAGAIN, but with the added bonus that we can't
- * expect write_space() to call us back to tell us
- * when to try sending again. We use the
- * SOCK_NOSPACE flag to diagnose... */
-
- LASSERT(rc != -ENOMEM);
-
- if (rc == 0 || rc == -EAGAIN) {
- if (test_bit(SOCK_NOSPACE,
- &conn->ksnc_sock->flags)) {
- rc = -EAGAIN;
- } else {
- static int counter;
-
- counter++;
- if ((counter & (-counter)) == counter)
- CWARN("%d ENOMEM tx %p\n",
- counter, conn);
- rc = -ENOMEM;
- }
+ if (rc <= 0) { /* Didn't write anything? */
+ unsigned long flags;
+ ksock_sched_t *sched;
+
+ if (rc == 0) /* some stacks return 0 instead of -EAGAIN */
+ rc = -EAGAIN;
+
+ if (rc != -EAGAIN)
+ break;
+
+ /* Check if EAGAIN is due to memory pressure */
+
+ sched = conn->ksnc_scheduler;
+ spin_lock_irqsave(&sched->kss_lock, flags);
+
+ if (!SOCK_TEST_NOSPACE(conn->ksnc_sock) &&
+ !conn->ksnc_tx_ready) {
+ /* SOCK_NOSPACE is set when the socket fills
+ * and cleared in the write_space callback
+ * (which also sets ksnc_tx_ready). If
+ * SOCK_NOSPACE and ksnc_tx_ready are BOTH
+ * zero, I didn't fill the socket and
+ * write_space won't reschedule me, so I
+ * return -ENOMEM to get my caller to retry
+ * after a timeout */
+ rc = -ENOMEM;
}
+
+ spin_unlock_irqrestore(&sched->kss_lock, flags);
break;
}
return (rc);
}
-void
-ksocknal_eager_ack (ksock_conn_t *conn)
-{
- int opt = 1;
- mm_segment_t oldmm = get_fs();
- struct socket *sock = conn->ksnc_sock;
-
- /* Remind the socket to ACK eagerly. If I don't, the socket might
- * think I'm about to send something it could piggy-back the ACK
- * on, introducing delay in completing zero-copy sends in my
- * peer. */
-
- set_fs(KERNEL_DS);
- sock->ops->setsockopt (sock, SOL_TCP, TCP_QUICKACK,
- (char *)&opt, sizeof (opt));
- set_fs(oldmm);
-}
-
int
ksocknal_recv_iov (ksock_conn_t *conn)
-{
+{
struct iovec *iov = conn->ksnc_rx_iov;
- int fragsize = iov->iov_len;
- unsigned long vaddr = (unsigned long)iov->iov_base;
- struct iovec fragiov = { .iov_base = (void *)vaddr,
- .iov_len = fragsize};
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &fragiov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = 0
- };
- mm_segment_t oldmm = get_fs();
- int rc;
-
- /* NB we can't trust socket ops to either consume our iovs
- * or leave them alone, so we only receive 1 frag at a time. */
+ int nob;
+ int rc;
+
LASSERT (conn->ksnc_rx_niov > 0);
- LASSERT (fragsize <= conn->ksnc_rx_nob_wanted);
- set_fs (KERNEL_DS);
- rc = sock_recvmsg (conn->ksnc_sock, &msg, fragsize, MSG_DONTWAIT);
- /* NB this is just a boolean............................^ */
- set_fs (oldmm);
+ /* Never touch conn->ksnc_rx_iov or change connection
+ * status inside ksocknal_lib_recv_iov */
+ rc = ksocknal_lib_recv_iov(conn);
- if (rc <= 0)
- return (rc);
+ if (rc <= 0)
+ return (rc);
- /* received something... */
- conn->ksnc_peer->ksnp_last_alive = jiffies;
- conn->ksnc_rx_deadline = jiffies +
- ksocknal_tunables.ksnd_io_timeout * HZ;
- mb(); /* order with setting rx_started */
- conn->ksnc_rx_started = 1;
+ /* received something... */
+ nob = rc;
+
+ conn->ksnc_peer->ksnp_last_alive = cfs_time_current();
+ conn->ksnc_rx_deadline = cfs_time_shift (ksocknal_tunables.ksnd_io_timeout);
+ mb(); /* order with setting rx_started */
+ conn->ksnc_rx_started = 1;
+
+ conn->ksnc_rx_nob_wanted -= nob;
+ conn->ksnc_rx_nob_left -= nob;
- conn->ksnc_rx_nob_wanted -= rc;
- conn->ksnc_rx_nob_left -= rc;
-
- if (rc < fragsize) {
- iov->iov_base = (void *)(vaddr + rc);
- iov->iov_len = fragsize - rc;
- return (-EAGAIN);
- }
+ do {
+ LASSERT (conn->ksnc_rx_niov > 0);
- conn->ksnc_rx_iov++;
- conn->ksnc_rx_niov--;
- return (1);
+ if (nob < iov->iov_len) {
+ iov->iov_len -= nob;
+ iov->iov_base = (void *)(((unsigned long)iov->iov_base) + nob);
+ return (-EAGAIN);
+ }
+
+ nob -= iov->iov_len;
+ conn->ksnc_rx_iov = ++iov;
+ conn->ksnc_rx_niov--;
+ } while (nob != 0);
+
+ return (rc);
}
int
ksocknal_recv_kiov (ksock_conn_t *conn)
{
ptl_kiov_t *kiov = conn->ksnc_rx_kiov;
- struct page *page = kiov->kiov_page;
- int offset = kiov->kiov_offset;
- int fragsize = kiov->kiov_len;
- unsigned long vaddr = ((unsigned long)kmap (page)) + offset;
- struct iovec fragiov = { .iov_base = (void *)vaddr,
- .iov_len = fragsize};
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &fragiov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = 0
- };
- mm_segment_t oldmm = get_fs();
- int rc;
-
- /* NB we can't trust socket ops to either consume our iovs
- * or leave them alone, so we only receive 1 frag at a time. */
- LASSERT (fragsize <= conn->ksnc_rx_nob_wanted);
+ int nob;
+ int rc;
LASSERT (conn->ksnc_rx_nkiov > 0);
- LASSERT (offset + fragsize <= PAGE_SIZE);
- set_fs (KERNEL_DS);
- rc = sock_recvmsg (conn->ksnc_sock, &msg, fragsize, MSG_DONTWAIT);
- /* NB this is just a boolean............................^ */
- set_fs (oldmm);
-
- kunmap (page);
+ /* Never touch conn->ksnc_rx_kiov or change connection
+ * status inside ksocknal_lib_recv_iov */
+ rc = ksocknal_lib_recv_kiov(conn);
- if (rc <= 0)
- return (rc);
+ if (rc <= 0)
+ return (rc);
- /* received something... */
- conn->ksnc_peer->ksnp_last_alive = jiffies;
- conn->ksnc_rx_deadline = jiffies +
- ksocknal_tunables.ksnd_io_timeout * HZ;
- mb(); /* order with setting rx_started */
+ /* received something... */
+ nob = rc;
+
+ conn->ksnc_peer->ksnp_last_alive = cfs_time_current();
+ conn->ksnc_rx_deadline = cfs_time_shift (ksocknal_tunables.ksnd_io_timeout);
+ mb(); /* order with setting rx_started */
conn->ksnc_rx_started = 1;
- conn->ksnc_rx_nob_wanted -= rc;
- conn->ksnc_rx_nob_left -= rc;
-
- if (rc < fragsize) {
- kiov->kiov_offset = offset + rc;
- kiov->kiov_len = fragsize - rc;
- return (-EAGAIN);
- }
+ conn->ksnc_rx_nob_wanted -= nob;
+ conn->ksnc_rx_nob_left -= nob;
+
+ do {
+ LASSERT (conn->ksnc_rx_nkiov > 0);
+
+ if (nob < kiov->kiov_len) {
+ kiov->kiov_offset += nob;
+ kiov->kiov_len -= nob;
+ return -EAGAIN;
+ }
- conn->ksnc_rx_kiov++;
- conn->ksnc_rx_nkiov--;
- return (1);
+ nob -= kiov->kiov_len;
+ conn->ksnc_rx_kiov = ++kiov;
+ conn->ksnc_rx_nkiov--;
+ } while (nob != 0);
+
+ return 1;
}
int
if (ksocknal_data.ksnd_stall_rx != 0) {
set_current_state (TASK_UNINTERRUPTIBLE);
- schedule_timeout (ksocknal_data.ksnd_stall_rx * HZ);
+ schedule_timeout(cfs_time_seconds (ksocknal_data.ksnd_stall_rx));
}
rc = ksocknal_getconnsock (conn);
(conn->ksnc_rx_state == SOCKNAL_RX_BODY ||
conn->ksnc_rx_state == SOCKNAL_RX_BODY_FWD)) {
/* Remind the socket to ack eagerly... */
- ksocknal_eager_ack(conn);
+ ksocknal_lib_eager_ack(conn);
}
rc = 1;
break;
spin_lock_irqsave (&sched->kss_lock, flags);
list_add_tail (&tx->tx_list, &sched->kss_zctxdone_list);
- wake_up (&sched->kss_waitq);
+ cfs_waitq_signal (&sched->kss_waitq);
spin_unlock_irqrestore (&sched->kss_lock, flags);
EXIT;
return (rc);
if (rc == -ENOMEM) {
+ static int counter;
+
+ counter++; /* exponential backoff warnings */
+ if ((counter & (-counter)) == counter)
+ CWARN("%d ENOMEM tx %p\n", counter, conn);
+
/* Queue on ksnd_enomem_conns for retry after a timeout */
spin_lock_irqsave(&ksocknal_data.ksnd_reaper_lock, flags);
LASSERT (conn->ksnc_tx_scheduled);
list_add_tail(&conn->ksnc_tx_list,
&ksocknal_data.ksnd_enomem_conns);
- if (!time_after_eq(jiffies + SOCKNAL_ENOMEM_RETRY,
+ if (!cfs_time_aftereq(cfs_time_add(cfs_time_current(),
+ SOCKNAL_ENOMEM_RETRY),
ksocknal_data.ksnd_reaper_waketime))
- wake_up (&ksocknal_data.ksnd_reaper_waitq);
+ cfs_waitq_signal (&ksocknal_data.ksnd_reaper_waitq);
spin_unlock_irqrestore(&ksocknal_data.ksnd_reaper_lock, flags);
return (rc);
/* Actual error */
LASSERT (rc < 0);
- if (!conn->ksnc_closing)
+ if (!conn->ksnc_closing) {
+ switch (rc) {
+ case -ECONNRESET:
+ LCONSOLE_WARN("Host %u.%u.%u.%u reset our connection "
+ "while we were sending data; it may have "
+ "rebooted.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
+ break;
+ default:
+ LCONSOLE_WARN("There was an unexpected network error "
+ "while writing to %u.%u.%u.%u: %d.\n",
+ HIPQUAD(conn->ksnc_ipaddr), rc);
+ break;
+ }
CERROR("[%p] Error %d on write to "LPX64
" ip %d.%d.%d.%d:%d\n", conn, rc,
conn->ksnc_peer->ksnp_nid,
HIPQUAD(conn->ksnc_ipaddr),
conn->ksnc_port);
+ }
ksocknal_close_conn_and_siblings (conn, rc);
ksocknal_tx_launched (tx);
list_add_tail (&route->ksnr_connect_list,
&ksocknal_data.ksnd_autoconnectd_routes);
- wake_up (&ksocknal_data.ksnd_autoconnectd_waitq);
+ cfs_waitq_signal (&ksocknal_data.ksnd_autoconnectd_waitq);
spin_unlock_irqrestore (&ksocknal_data.ksnd_autoconnectd_lock, flags);
}
const int nob = 0;
#else
int nob = atomic_read(&c->ksnc_tx_nob) +
- c->ksnc_sock->sk->sk_wmem_queued;
+ SOCK_WMEM_QUEUED(c->ksnc_sock);
#endif
LASSERT (!c->ksnc_closing);
spin_lock_irqsave (&sched->kss_lock, flags);
if (list_empty(&conn->ksnc_tx_queue) &&
- conn->ksnc_sock->sk->sk_wmem_queued == 0) {
+ SOCK_WMEM_QUEUED(conn->ksnc_sock) == 0) {
/* First packet starts the timeout */
- conn->ksnc_tx_deadline = jiffies +
- ksocknal_tunables.ksnd_io_timeout * HZ;
+ conn->ksnc_tx_deadline = cfs_time_shift(ksocknal_tunables.ksnd_io_timeout);
conn->ksnc_tx_bufnob = 0;
mb(); /* order with adding to tx_queue */
}
list_add_tail (&conn->ksnc_tx_list,
&sched->kss_tx_conns);
conn->ksnc_tx_scheduled = 1;
- wake_up (&sched->kss_waitq);
+ cfs_waitq_signal (&sched->kss_waitq);
}
spin_unlock_irqrestore (&sched->kss_lock, flags);
continue;
/* too soon to retry this guy? */
- if (!time_after_eq (jiffies, route->ksnr_timeout))
+ if (!cfs_time_aftereq (cfs_time_current(), route->ksnr_timeout))
continue;
return (route);
int
ksocknal_thread_start (int (*fn)(void *arg), void *arg)
{
- long pid = kernel_thread (fn, arg, 0);
+ long pid = cfs_kernel_thread (fn, arg, 0);
unsigned long flags;
if (pid < 0)
spin_lock_irqsave (&sched->kss_lock, flags);
list_add_tail (&conn->ksnc_rx_list, &sched->kss_rx_conns);
- wake_up (&sched->kss_waitq);
+ cfs_waitq_signal (&sched->kss_waitq);
spin_unlock_irqrestore (&sched->kss_lock, flags);
}
LASSERT (conn->ksnc_rx_state == SOCKNAL_RX_GET_FMB);
LASSERT (kpr_routing(&ksocknal_data.ksnd_router));
- if (payload_nob <= SOCKNAL_SMALL_FWD_PAGES * PAGE_SIZE)
+ if (payload_nob <= SOCKNAL_SMALL_FWD_PAGES * CFS_PAGE_SIZE)
pool = &ksocknal_data.ksnd_small_fmp;
else
pool = &ksocknal_data.ksnd_large_fmp;
LASSERT (conn->ksnc_rx_state == SOCKNAL_RX_GET_FMB);
LASSERT (conn->ksnc_rx_nob_wanted == conn->ksnc_rx_nob_left);
LASSERT (payload_nob >= 0);
- LASSERT (payload_nob <= fmb->fmb_pool->fmp_buff_pages * PAGE_SIZE);
- LASSERT (sizeof (ptl_hdr_t) < PAGE_SIZE);
+ LASSERT (payload_nob <= fmb->fmb_pool->fmp_buff_pages * CFS_PAGE_SIZE);
+ LASSERT (sizeof (ptl_hdr_t) < CFS_PAGE_SIZE);
LASSERT (fmb->fmb_kiov[0].kiov_offset == 0);
/* Take a ref on the conn's peer to prevent module unload before
while (nob > 0) {
LASSERT (niov < fmb->fmb_pool->fmp_buff_pages);
LASSERT (fmb->fmb_kiov[niov].kiov_offset == 0);
- fmb->fmb_kiov[niov].kiov_len = MIN (PAGE_SIZE, nob);
- nob -= PAGE_SIZE;
+ fmb->fmb_kiov[niov].kiov_len = MIN (CFS_PAGE_SIZE, nob);
+ nob -= CFS_PAGE_SIZE;
niov++;
}
return (0);
}
-void
-ksocknal_data_ready (struct sock *sk, int n)
+/*
+ * Add connection to kss_rx_conns of scheduler
+ * and wakeup the scheduler.
+ */
+void ksocknal_read_callback (ksock_conn_t *conn)
{
+ ksock_sched_t *sched;
unsigned long flags;
- ksock_conn_t *conn;
- ksock_sched_t *sched;
ENTRY;
- /* interleave correctly with closing sockets... */
- read_lock (&ksocknal_data.ksnd_global_lock);
+ sched = conn->ksnc_scheduler;
- conn = sk->sk_user_data;
- if (conn == NULL) { /* raced with ksocknal_terminate_conn */
- LASSERT (sk->sk_data_ready != &ksocknal_data_ready);
- sk->sk_data_ready (sk, n);
- } else {
- sched = conn->ksnc_scheduler;
-
- spin_lock_irqsave (&sched->kss_lock, flags);
+ spin_lock_irqsave (&sched->kss_lock, flags);
- conn->ksnc_rx_ready = 1;
+ conn->ksnc_rx_ready = 1;
- if (!conn->ksnc_rx_scheduled) { /* not being progressed */
- list_add_tail(&conn->ksnc_rx_list,
- &sched->kss_rx_conns);
- conn->ksnc_rx_scheduled = 1;
- /* extra ref for scheduler */
- atomic_inc (&conn->ksnc_refcount);
+ if (!conn->ksnc_rx_scheduled) { /* not being progressed */
+ list_add_tail(&conn->ksnc_rx_list,
+ &sched->kss_rx_conns);
+ conn->ksnc_rx_scheduled = 1;
+ /* extra ref for scheduler */
+ atomic_inc (&conn->ksnc_refcount);
- wake_up (&sched->kss_waitq);
- }
-
- spin_unlock_irqrestore (&sched->kss_lock, flags);
- }
-
- read_unlock (&ksocknal_data.ksnd_global_lock);
+ cfs_waitq_signal (&sched->kss_waitq);
+ }
+ spin_unlock_irqrestore (&sched->kss_lock, flags);
EXIT;
-}
+}
-void
-ksocknal_write_space (struct sock *sk)
-{
+/*
+ * Add connection to kss_tx_conns of scheduler
+ * and wakeup the scheduler.
+ */
+void ksocknal_write_callback (ksock_conn_t *conn)
+{
+ ksock_sched_t *sched;
unsigned long flags;
- ksock_conn_t *conn;
- ksock_sched_t *sched;
-
- /* interleave correctly with closing sockets... */
- read_lock (&ksocknal_data.ksnd_global_lock);
-
- conn = sk->sk_user_data;
-
- CDEBUG(D_NET, "sk %p wspace %d low water %d conn %p%s%s%s\n",
- sk, tcp_wspace(sk), SOCKNAL_TX_LOW_WATER(sk), conn,
- (conn == NULL) ? "" : (conn->ksnc_tx_ready ?
- " ready" : " blocked"),
- (conn == NULL) ? "" : (conn->ksnc_tx_scheduled ?
- " scheduled" : " idle"),
- (conn == NULL) ? "" : (list_empty (&conn->ksnc_tx_queue) ?
- " empty" : " queued"));
-
- if (conn == NULL) { /* raced with ksocknal_terminate_conn */
- LASSERT (sk->sk_write_space != &ksocknal_write_space);
- sk->sk_write_space (sk);
-
- read_unlock (&ksocknal_data.ksnd_global_lock);
- return;
- }
-
- if (tcp_wspace(sk) >= SOCKNAL_TX_LOW_WATER(sk)) { /* got enough space */
- clear_bit (SOCK_NOSPACE, &sk->sk_socket->flags);
-
- sched = conn->ksnc_scheduler;
+ ENTRY;
+
+ sched = conn->ksnc_scheduler;
- spin_lock_irqsave (&sched->kss_lock, flags);
+ spin_lock_irqsave (&sched->kss_lock, flags);
- conn->ksnc_tx_ready = 1;
+ conn->ksnc_tx_ready = 1;
- if (!conn->ksnc_tx_scheduled && // not being progressed
- !list_empty(&conn->ksnc_tx_queue)){//packets to send
- list_add_tail (&conn->ksnc_tx_list,
- &sched->kss_tx_conns);
- conn->ksnc_tx_scheduled = 1;
- /* extra ref for scheduler */
- atomic_inc (&conn->ksnc_refcount);
+ if (!conn->ksnc_tx_scheduled && // not being progressed
+ !list_empty(&conn->ksnc_tx_queue)){//packets to send
+ list_add_tail (&conn->ksnc_tx_list,
+ &sched->kss_tx_conns);
+ conn->ksnc_tx_scheduled = 1;
+ /* extra ref for scheduler */
+ atomic_inc (&conn->ksnc_refcount);
- wake_up (&sched->kss_waitq);
- }
+ cfs_waitq_signal (&sched->kss_waitq);
+ }
- spin_unlock_irqrestore (&sched->kss_lock, flags);
- }
+ spin_unlock_irqrestore (&sched->kss_lock, flags);
- read_unlock (&ksocknal_data.ksnd_global_lock);
+ EXIT;
}
int
ksocknal_sock_write (struct socket *sock, void *buffer, int nob)
{
- int rc;
- mm_segment_t oldmm = get_fs();
-
- while (nob > 0) {
- struct iovec iov = {
- .iov_base = buffer,
- .iov_len = nob
- };
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &iov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = 0
- };
-
- set_fs (KERNEL_DS);
- rc = sock_sendmsg (sock, &msg, iov.iov_len);
- set_fs (oldmm);
-
- if (rc < 0)
- return (rc);
-
- if (rc == 0) {
- CERROR ("Unexpected zero rc\n");
- return (-ECONNABORTED);
- }
-
- buffer = ((char *)buffer) + rc;
- nob -= rc;
- }
-
- return (0);
+ return ksocknal_lib_sock_write(sock, buffer, nob);
}
int
ksocknal_sock_read (struct socket *sock, void *buffer, int nob)
{
- int rc;
- mm_segment_t oldmm = get_fs();
-
- while (nob > 0) {
- struct iovec iov = {
- .iov_base = buffer,
- .iov_len = nob
- };
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &iov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = 0
- };
-
- set_fs (KERNEL_DS);
- rc = sock_recvmsg (sock, &msg, iov.iov_len, 0);
- set_fs (oldmm);
-
- if (rc < 0)
- return (rc);
-
- if (rc == 0)
- return (-ECONNABORTED);
-
- buffer = ((char *)buffer) + rc;
- nob -= rc;
- }
-
- return (0);
+ return ksocknal_lib_sock_read(sock, buffer, nob);
}
int
int type;
ptl_hdr_t hdr;
ptl_magicversion_t *hmv;
+ char ipbuf[PTL_NALFMT_SIZE];
hmv = (ptl_magicversion_t *)&hdr.dest_nid;
LASSERT (sizeof (*hmv) == sizeof (hdr.dest_nid));
if (*nid == PTL_NID_ANY) { /* don't know peer's nid yet */
*nid = le64_to_cpu(hdr.src_nid);
} else if (*nid != le64_to_cpu (hdr.src_nid)) {
+ LCONSOLE_ERROR("Connected successfully to nid "LPX64" on host "
+ "%u.%u.%u.%u, but they claimed they were nid "
+ LPX64" (%s); please check your Lustre "
+ "configuration.\n",
+ *nid, HIPQUAD(conn->ksnc_ipaddr),
+ le64_to_cpu(hdr.src_nid),
+ portals_nid2str(SOCKNAL,
+ le64_to_cpu(hdr.src_nid),
+ ipbuf));
+
CERROR ("Connected to nid "LPX64"@%u.%u.%u.%u "
"but expecting "LPX64"\n",
le64_to_cpu (hdr.src_nid),
int
ksocknal_get_conn_tunables (ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
{
- mm_segment_t oldmm = get_fs ();
- struct socket *sock = conn->ksnc_sock;
- int len;
- int rc;
-
- rc = ksocknal_getconnsock (conn);
- if (rc != 0) {
- LASSERT (conn->ksnc_closing);
- *txmem = *rxmem = *nagle = 0;
- return (-ESHUTDOWN);
- }
-
- set_fs (KERNEL_DS);
-
- len = sizeof(*txmem);
- rc = sock_getsockopt(sock, SOL_SOCKET, SO_SNDBUF,
- (char *)txmem, &len);
- if (rc == 0) {
- len = sizeof(*rxmem);
- rc = sock_getsockopt(sock, SOL_SOCKET, SO_RCVBUF,
- (char *)rxmem, &len);
- }
- if (rc == 0) {
- len = sizeof(*nagle);
- rc = sock->ops->getsockopt(sock, SOL_TCP, TCP_NODELAY,
- (char *)nagle, &len);
- }
-
- set_fs (oldmm);
- ksocknal_putconnsock (conn);
-
- if (rc == 0)
- *nagle = !*nagle;
- else
- *txmem = *rxmem = *nagle = 0;
-
- return (rc);
-}
-
-int
-ksocknal_setup_sock (struct socket *sock)
-{
- mm_segment_t oldmm = get_fs ();
- int rc;
- int option;
- int keep_idle;
- int keep_intvl;
- int keep_count;
- int do_keepalive;
- struct linger linger;
-
- sock->sk->sk_allocation = GFP_NOFS;
-
- /* Ensure this socket aborts active sends immediately when we close
- * it. */
-
- linger.l_onoff = 0;
- linger.l_linger = 0;
-
- set_fs (KERNEL_DS);
- rc = sock_setsockopt (sock, SOL_SOCKET, SO_LINGER,
- (char *)&linger, sizeof (linger));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set SO_LINGER: %d\n", rc);
- return (rc);
- }
-
- option = -1;
- set_fs (KERNEL_DS);
- rc = sock->ops->setsockopt (sock, SOL_TCP, TCP_LINGER2,
- (char *)&option, sizeof (option));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set SO_LINGER2: %d\n", rc);
- return (rc);
- }
-
- if (!ksocknal_tunables.ksnd_nagle) {
- option = 1;
-
- set_fs (KERNEL_DS);
- rc = sock->ops->setsockopt (sock, SOL_TCP, TCP_NODELAY,
- (char *)&option, sizeof (option));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't disable nagle: %d\n", rc);
- return (rc);
- }
- }
-
- if (ksocknal_tunables.ksnd_buffer_size > 0) {
- option = ksocknal_tunables.ksnd_buffer_size;
-
- set_fs (KERNEL_DS);
- rc = sock_setsockopt (sock, SOL_SOCKET, SO_SNDBUF,
- (char *)&option, sizeof (option));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set send buffer %d: %d\n",
- option, rc);
- return (rc);
- }
-
- set_fs (KERNEL_DS);
- rc = sock_setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
- (char *)&option, sizeof (option));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set receive buffer %d: %d\n",
- option, rc);
- return (rc);
- }
- }
-
- /* snapshot tunables */
- keep_idle = ksocknal_tunables.ksnd_keepalive_idle;
- keep_count = ksocknal_tunables.ksnd_keepalive_count;
- keep_intvl = ksocknal_tunables.ksnd_keepalive_intvl;
-
- do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
-
- option = (do_keepalive ? 1 : 0);
- set_fs (KERNEL_DS);
- rc = sock_setsockopt (sock, SOL_SOCKET, SO_KEEPALIVE,
- (char *)&option, sizeof (option));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
- return (rc);
- }
-
- if (!do_keepalive)
- return (0);
-
- set_fs (KERNEL_DS);
- rc = sock->ops->setsockopt (sock, SOL_TCP, TCP_KEEPIDLE,
- (char *)&keep_idle, sizeof (keep_idle));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set TCP_KEEPIDLE: %d\n", rc);
- return (rc);
- }
-
- set_fs (KERNEL_DS);
- rc = sock->ops->setsockopt (sock, SOL_TCP, TCP_KEEPINTVL,
- (char *)&keep_intvl, sizeof (keep_intvl));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set TCP_KEEPINTVL: %d\n", rc);
- return (rc);
- }
-
- set_fs (KERNEL_DS);
- rc = sock->ops->setsockopt (sock, SOL_TCP, TCP_KEEPCNT,
- (char *)&keep_count, sizeof (keep_count));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set TCP_KEEPCNT: %d\n", rc);
- return (rc);
- }
-
- return (0);
+ return ksocknal_lib_get_conn_tunables(conn, txmem, rxmem, nagle);
}
int
ksocknal_connect_peer (ksock_route_t *route, int type)
{
- struct sockaddr_in ipaddr;
- mm_segment_t oldmm = get_fs();
- struct timeval tv;
- int fd;
struct socket *sock;
int rc;
+ int port;
+ int may_retry;
- rc = sock_create (PF_INET, SOCK_STREAM, 0, &sock);
- if (rc != 0) {
- CERROR ("Can't create autoconnect socket: %d\n", rc);
- return (rc);
- }
-
- /* Ugh; have to map_fd for compatibility with sockets passed in
- * from userspace. And we actually need the sock->file refcounting
- * that this gives you :) */
-
- fd = sock_map_fd (sock);
- if (fd < 0) {
- sock_release (sock);
- CERROR ("sock_map_fd error %d\n", fd);
- return (fd);
- }
+ /* Iterate through reserved ports. When typed connections are
+ * used, we will need to bind to multiple ports, but we only know
+ * this at connect time. But, by that time we've already called
+ * bind() so we need a new socket. */
- /* NB the fd now owns the ref on sock->file */
- LASSERT (sock->file != NULL);
- LASSERT (file_count(sock->file) == 1);
+ for (port = 1023; port > 512; --port) {
- /* Set the socket timeouts, so our connection attempt completes in
- * finite time */
- tv.tv_sec = ksocknal_tunables.ksnd_io_timeout;
- tv.tv_usec = 0;
+ rc = ksocknal_lib_connect_sock(&sock, &may_retry, route, port);
- set_fs (KERNEL_DS);
- rc = sock_setsockopt (sock, SOL_SOCKET, SO_SNDTIMEO,
- (char *)&tv, sizeof (tv));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set send timeout %d: %d\n",
- ksocknal_tunables.ksnd_io_timeout, rc);
- goto out;
- }
-
- set_fs (KERNEL_DS);
- rc = sock_setsockopt (sock, SOL_SOCKET, SO_RCVTIMEO,
- (char *)&tv, sizeof (tv));
- set_fs (oldmm);
- if (rc != 0) {
- CERROR ("Can't set receive timeout %d: %d\n",
- ksocknal_tunables.ksnd_io_timeout, rc);
- goto out;
- }
-
- if (route->ksnr_myipaddr != 0) {
- /* Bind to the local IP address */
- memset (&ipaddr, 0, sizeof (ipaddr));
- ipaddr.sin_family = AF_INET;
- ipaddr.sin_port = htons (0); /* ANY */
- ipaddr.sin_addr.s_addr = htonl(route->ksnr_myipaddr);
-
- rc = sock->ops->bind (sock, (struct sockaddr *)&ipaddr,
- sizeof (ipaddr));
- if (rc != 0) {
- CERROR ("Can't bind to local IP %u.%u.%u.%u: %d\n",
- HIPQUAD(route->ksnr_myipaddr), rc);
- goto out;
+ if (rc == 0) {
+ rc = ksocknal_create_conn(route, sock, type);
+ cfs_put_file(KSN_SOCK2FILE(sock));
+ return rc;
}
- }
-
- memset (&ipaddr, 0, sizeof (ipaddr));
- ipaddr.sin_family = AF_INET;
- ipaddr.sin_port = htons (route->ksnr_port);
- ipaddr.sin_addr.s_addr = htonl (route->ksnr_ipaddr);
-
- rc = sock->ops->connect (sock, (struct sockaddr *)&ipaddr,
- sizeof (ipaddr), sock->file->f_flags);
- if (rc != 0) {
- CERROR ("Can't connect to nid "LPX64
- " local IP: %u.%u.%u.%u,"
- " remote IP: %u.%u.%u.%u/%d: %d\n",
- route->ksnr_peer->ksnp_nid,
- HIPQUAD(route->ksnr_myipaddr),
- HIPQUAD(route->ksnr_ipaddr),
- route->ksnr_port, rc);
- goto out;
- }
- rc = ksocknal_create_conn (route, sock, type);
- if (rc == 0) {
- /* Take an extra ref on sock->file to compensate for the
- * upcoming close which will lose fd's ref on it. */
- get_file (sock->file);
+ if (!may_retry)
+ return rc;
}
- out:
- sys_close (fd);
- return (rc);
+ CERROR("Out of ports trying to bind to a reserved port\n");
+ return (-EADDRINUSE);
}
void
ksocknal_autoconnect (ksock_route_t *route)
{
- LIST_HEAD (zombies);
+ CFS_LIST_HEAD (zombies);
ksock_tx_t *tx;
ksock_peer_t *peer;
unsigned long flags;
int rc;
int type;
+ char *err_msg = NULL;
for (;;) {
for (type = 0; type < SOCKNAL_CONN_NTYPES; type++)
LASSERT (type < SOCKNAL_CONN_NTYPES);
rc = ksocknal_connect_peer (route, type);
-
if (rc != 0)
break;
}
}
+ switch (rc) {
+ /* "normal" errors */
+ case -ECONNREFUSED:
+ LCONSOLE_ERROR("Connection was refused by host %u.%u.%u.%u on "
+ "port %d; check that Lustre is running on that "
+ "node.\n",
+ HIPQUAD(route->ksnr_ipaddr),
+ route->ksnr_port);
+ break;
+ case -EHOSTUNREACH:
+ case -ENETUNREACH:
+ LCONSOLE_ERROR("Host %u.%u.%u.%u was unreachable; the network "
+ "or that node may be down, or Lustre may be "
+ "misconfigured.\n",
+ HIPQUAD(route->ksnr_ipaddr));
+ break;
+ case -ETIMEDOUT:
+ LCONSOLE_ERROR("Connecting to host %u.%u.%u.%u on port %d took "
+ "too long; that node may be hung or "
+ "experiencing high load.\n",
+ HIPQUAD(route->ksnr_ipaddr),
+ route->ksnr_port);
+ break;
+ /* errors that should be rare */
+ case -EPROTO:
+ err_msg = "Portals could not negotiate a connection";
+ break;
+ case -EAGAIN:
+ case -EADDRINUSE:
+ /* -EAGAIN is out of ports, but we specify the ports
+ * manually. we really should never get this */
+ err_msg = "no privileged ports were available";
+ break;
+ default:
+ err_msg = "unknown error";
+ break;
+ }
+
+ if (err_msg) {
+ LCONSOLE_ERROR("There was an unexpected error connecting to host "
+ "%u.%u.%u.%u on port %d: %s (error code %d).\n",
+ HIPQUAD(route->ksnr_ipaddr),
+ route->ksnr_port,
+ err_msg, -rc);
+ }
+
/* Connection attempt failed */
write_lock_irqsave (&ksocknal_data.ksnd_global_lock, flags);
/* This is a retry rather than a new connection */
LASSERT (route->ksnr_retry_interval != 0);
- route->ksnr_timeout = jiffies + route->ksnr_retry_interval;
+ route->ksnr_timeout = cfs_time_add(cfs_time_current(),
+ route->ksnr_retry_interval);
route->ksnr_retry_interval = MIN (route->ksnr_retry_interval * 2,
SOCKNAL_MAX_RECONNECT_INTERVAL);
/* Don't need the {get,put}connsock dance to deref ksnc_sock... */
LASSERT (!conn->ksnc_closing);
- if (conn->ksnc_sock->sk->sk_err != 0) {
- /* Something (e.g. failed keepalive) set the socket error */
+ if (SOCK_ERROR(conn->ksnc_sock) != 0) {
atomic_inc (&conn->ksnc_refcount);
+
+ switch (SOCK_ERROR(conn->ksnc_sock)) {
+ case ECONNRESET:
+ LCONSOLE_WARN("A connection with %u.%u.%u.%u "
+ "was reset; they may have "
+ "rebooted.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
+ break;
+ case ETIMEDOUT:
+ LCONSOLE_WARN("A connection with %u.%u.%u.%u "
+ "timed out; the network or that "
+ "node may be down.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
+ break;
+ default:
+ LCONSOLE_WARN("An unexpected network error "
+ "occurred with %u.%u.%u.%u: %d.\n",
+ HIPQUAD(conn->ksnc_ipaddr),
+ SOCK_ERROR(conn->ksnc_sock));
+ break;
+ }
+
+ /* Something (e.g. failed keepalive) set the socket error */
CERROR ("Socket error %d: "LPX64" %p %d.%d.%d.%d\n",
- conn->ksnc_sock->sk->sk_err, peer->ksnp_nid,
+ SOCK_ERROR(conn->ksnc_sock), peer->ksnp_nid,
conn, HIPQUAD(conn->ksnc_ipaddr));
+
return (conn);
}
if (conn->ksnc_rx_started &&
- time_after_eq (jiffies, conn->ksnc_rx_deadline)) {
+ cfs_time_aftereq (cfs_time_current(),
+ conn->ksnc_rx_deadline)) {
/* Timed out incomplete incoming message */
atomic_inc (&conn->ksnc_refcount);
+ LCONSOLE_ERROR("A timeout occurred receiving data from "
+ "%u.%u.%u.%u; the network or that node "
+ "may be down.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
CERROR ("Timed out RX from "LPX64" %p %d.%d.%d.%d\n",
peer->ksnp_nid,conn,HIPQUAD(conn->ksnc_ipaddr));
return (conn);
}
if ((!list_empty (&conn->ksnc_tx_queue) ||
- conn->ksnc_sock->sk->sk_wmem_queued != 0) &&
- time_after_eq (jiffies, conn->ksnc_tx_deadline)) {
+ SOCK_WMEM_QUEUED(conn->ksnc_sock) != 0) &&
+ cfs_time_aftereq (cfs_time_current(),
+ conn->ksnc_tx_deadline)) {
/* Timed out messages queued for sending or
* buffered in the socket's send buffer */
atomic_inc (&conn->ksnc_refcount);
+ LCONSOLE_ERROR("A timeout occurred sending data to "
+ "%u.%u.%u.%u; the network or that node "
+ "may be down.\n",
+ HIPQUAD(conn->ksnc_ipaddr));
CERROR ("Timed out TX to "LPX64" %s%d %p %d.%d.%d.%d\n",
peer->ksnp_nid,
list_empty (&conn->ksnc_tx_queue) ? "" : "Q ",
- conn->ksnc_sock->sk->sk_wmem_queued, conn,
+ SOCK_WMEM_QUEUED(conn->ksnc_sock), conn,
HIPQUAD(conn->ksnc_ipaddr));
return (conn);
}
int
ksocknal_reaper (void *arg)
{
- wait_queue_t wait;
+ cfs_waitlink_t wait;
unsigned long flags;
ksock_conn_t *conn;
ksock_sched_t *sched;
struct list_head enomem_conns;
int nenomem_conns;
- int timeout;
+ cfs_duration_t timeout;
int i;
int peer_index = 0;
- unsigned long deadline = jiffies;
+ cfs_time_t deadline = cfs_time_current();
kportal_daemonize ("ksocknal_reaper");
kportal_blockallsigs ();
- INIT_LIST_HEAD(&enomem_conns);
- init_waitqueue_entry (&wait, current);
+ CFS_INIT_LIST_HEAD(&enomem_conns);
+ cfs_waitlink_init (&wait);
spin_lock_irqsave (&ksocknal_data.ksnd_reaper_lock, flags);
LASSERT (conn->ksnc_tx_scheduled);
conn->ksnc_tx_ready = 1;
list_add_tail (&conn->ksnc_tx_list, &sched->kss_tx_conns);
- wake_up (&sched->kss_waitq);
+ cfs_waitq_signal (&sched->kss_waitq);
spin_unlock_irqrestore (&sched->kss_lock, flags);
nenomem_conns++;
}
/* careful with the jiffy wrap... */
- while ((timeout = (int)(deadline - jiffies)) <= 0) {
+ while ((timeout = cfs_time_sub(deadline,
+ cfs_time_current())) <= 0) {
const int n = 4;
const int p = 1;
int chunk = ksocknal_data.ksnd_peer_hash_size;
ksocknal_data.ksnd_peer_hash_size;
}
- deadline += p * HZ;
+ deadline = cfs_time_add(deadline, cfs_time_seconds(p));
}
if (nenomem_conns != 0) {
* if any go back on my enomem list. */
timeout = SOCKNAL_ENOMEM_RETRY;
}
- ksocknal_data.ksnd_reaper_waketime = jiffies + timeout;
+ ksocknal_data.ksnd_reaper_waketime =
+ cfs_time_add(cfs_time_current(), timeout);
set_current_state (TASK_INTERRUPTIBLE);
- add_wait_queue (&ksocknal_data.ksnd_reaper_waitq, &wait);
+ cfs_waitq_add (&ksocknal_data.ksnd_reaper_waitq, &wait);
if (!ksocknal_data.ksnd_shuttingdown &&
list_empty (&ksocknal_data.ksnd_deathrow_conns) &&
list_empty (&ksocknal_data.ksnd_zombie_conns))
- schedule_timeout (timeout);
+ cfs_waitq_timedwait (&wait, timeout);
set_current_state (TASK_RUNNING);
- remove_wait_queue (&ksocknal_data.ksnd_reaper_waitq, &wait);
+ cfs_waitq_del (&ksocknal_data.ksnd_reaper_waitq, &wait);
spin_lock_irqsave (&ksocknal_data.ksnd_reaper_lock, flags);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff