/**
* Server side bulk abort. Idempotent. Not thread-safe (i.e. only
* serialises with completion callback)
*/
void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc)
{
struct l_wait_info lwi;
int rc;
LASSERT(!in_interrupt()); /* might sleep */
if (!ptlrpc_server_bulk_active(desc)) /* completed or */
return; /* never started */
/* We used to poison the pages with 0xab here because we did not want to
* send any meaningful data over the wire for evicted clients (bug 9297)
* However, this is no longer safe now that we use the page cache on the
* OSS (bug 20560) */
/* The unlink ensures the callback happens ASAP and is the last
* one. If it fails, it must be because completion just happened,
* but we must still l_wait_event() in this case, to give liblustre
* a chance to run server_bulk_callback()*/
mdunlink_iterate_helper(desc->bd_mds, desc->bd_md_max_brw);
for (;;) {
/* Network access will complete in finite time but the HUGE
* timeout lets us CWARN for visibility of sluggish NALs */
lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(LONG_UNLINK),
cfs_time_seconds(1), NULL, NULL);
rc = l_wait_event(desc->bd_waitq,
!ptlrpc_server_bulk_active(desc), &lwi);
if (rc == 0)
return;
LASSERT(rc == -ETIMEDOUT);
CWARN("Unexpectedly long timeout: desc %p\n", desc);
}
}
/**
* Disconnect a bulk desc from the network. Idempotent. Not
* thread-safe (i.e. only interlocks with completion callback).
* Returns 1 on success or 0 if network unregistration failed for whatever
* reason.
*/
int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async)
{
struct ptlrpc_bulk_desc *desc = req->rq_bulk;
wait_queue_head_t *wq;
struct l_wait_info lwi;
int rc;
LASSERT(!in_interrupt()); /* might sleep */
/* Let's setup deadline for reply unlink. */
if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
async && req->rq_bulk_deadline == 0 && cfs_fail_val == 0)
req->rq_bulk_deadline = ktime_get_real_seconds() + LONG_UNLINK;
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
return 1; /* never registered */
LASSERT(desc->bd_req == req); /* bd_req NULL until registered */
/* the unlink ensures the callback happens ASAP and is the last
* one. If it fails, it must be because completion just happened,
* but we must still l_wait_event() in this case to give liblustre
* a chance to run client_bulk_callback()
*/
mdunlink_iterate_helper(desc->bd_mds, desc->bd_md_max_brw);
if (ptlrpc_client_bulk_active(req) == 0) /* completed or */
return 1; /* never registered */
/* Move to "Unregistering" phase as bulk was not unlinked yet. */
ptlrpc_rqphase_move(req, RQ_PHASE_UNREG_BULK);
/* Do not wait for unlink to finish. */
if (async)
return 0;
if (req->rq_set)
wq = &req->rq_set->set_waitq;
else
wq = &req->rq_reply_waitq;
for (;;) {
/* Network access will complete in finite time but the HUGE
* timeout lets us CWARN for visibility of sluggish LNDs
*/
lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(LONG_UNLINK),
cfs_time_seconds(1), NULL, NULL);
rc = l_wait_event(*wq, !ptlrpc_client_bulk_active(req), &lwi);
if (rc == 0) {
ptlrpc_rqphase_move(req, req->rq_next_phase);
return 1;
}
LASSERT(rc == -ETIMEDOUT);
DEBUG_REQ(D_WARNING, req, "Unexpectedly long timeout: desc %p",
desc);
}
return 0;
}
/**
* Callback handler for receiving incoming glimpse ASTs.
*
* This only can happen on client side. After handling the glimpse AST
* we also consider dropping the lock here if it is unused locally for a
* long time.
*/
static void ldlm_handle_gl_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int rc = -ENOSYS;
LDLM_DEBUG(lock, "client glimpse AST callback handler");
if (lock->l_glimpse_ast != NULL)
rc = lock->l_glimpse_ast(lock, req);
if (req->rq_repmsg != NULL) {
ptlrpc_reply(req);
} else {
req->rq_status = rc;
ptlrpc_error(req);
}
lock_res_and_lock(lock);
if (lock->l_granted_mode == LCK_PW &&
!lock->l_readers && !lock->l_writers &&
cfs_time_after(cfs_time_current(),
cfs_time_add(lock->l_last_used,
cfs_time_seconds(10)))) {
unlock_res_and_lock(lock);
if (ldlm_bl_to_thread_lock(ns, NULL, lock))
ldlm_handle_bl_callback(ns, NULL, lock);
return;
}
unlock_res_and_lock(lock);
LDLM_LOCK_RELEASE(lock);
}
static int
proc_trigger_stack_reset(struct ctl_table *table, int write,
void __user *buffer, size_t *lenp, loff_t *ppos)
{
int rc = 0;
int i = 1;
kgn_device_t *dev;
ENTRY;
if (!write) {
/* read */
rc = proc_dointvec(table, write, buffer, lenp, ppos);
RETURN(rc);
}
/* only device 0 gets the handle, see kgnilnd_dev_init */
dev = &kgnilnd_data.kgn_devices[0];
LASSERTF(dev != NULL, "dev 0 is NULL\n");
kgnilnd_critical_error(dev->gnd_err_handle);
/* Wait for the reset to complete. This prevents any races in testing
* where we'd immediately try to send traffic again */
while (kgnilnd_data.kgn_needs_reset != 0) {
i++;
LCONSOLE((((i) & (-i)) == i) ? D_WARNING : D_NET,
"Waiting for stack reset request to clear\n");
cfs_pause(cfs_time_seconds(1 * i));
}
RETURN(rc);
}
int osc_object_is_contended(struct osc_object *obj)
{
struct osc_device *dev = lu2osc_dev(obj->oo_cl.co_lu.lo_dev);
int osc_contention_time = dev->od_contention_time;
unsigned long cur_time = cfs_time_current();
unsigned long retry_time;
if (OBD_FAIL_CHECK(OBD_FAIL_OSC_OBJECT_CONTENTION))
return 1;
if (!obj->oo_contended)
return 0;
/*
* I like copy-paste. the code is copied from
* ll_file_is_contended.
*/
retry_time = cfs_time_add(obj->oo_contention_time,
cfs_time_seconds(osc_contention_time));
if (cfs_time_after(cur_time, retry_time)) {
osc_object_clear_contended(obj);
return 0;
}
return 1;
}
static int proc_console_min_delay_cs(struct ctl_table *table, int write,
void __user *buffer, size_t *lenp,
loff_t *ppos)
{
int rc, min_delay_cs;
struct ctl_table dummy = *table;
long d;
dummy.data = &min_delay_cs;
dummy.proc_handler = &proc_dointvec;
if (!write) { /* read */
min_delay_cs = cfs_duration_sec(libcfs_console_min_delay * 100);
rc = proc_dointvec(&dummy, write, buffer, lenp, ppos);
return rc;
}
/* write */
min_delay_cs = 0;
rc = proc_dointvec(&dummy, write, buffer, lenp, ppos);
if (rc < 0)
return rc;
if (min_delay_cs <= 0)
return -EINVAL;
d = cfs_time_seconds(min_delay_cs) / 100;
if (d == 0 || d > libcfs_console_max_delay)
return -EINVAL;
libcfs_console_min_delay = d;
return rc;
}
void ptlrpc_ni_fini(void)
{
wait_queue_head_t waitq;
struct l_wait_info lwi;
int rc;
int retries;
/* Wait for the event queue to become idle since there may still be
* messages in flight with pending events (i.e. the fire-and-forget
* messages == client requests and "non-difficult" server
* replies */
for (retries = 0;; retries++) {
rc = LNetEQFree(ptlrpc_eq_h);
switch (rc) {
default:
LBUG();
case 0:
LNetNIFini();
return;
case -EBUSY:
if (retries != 0)
CWARN("Event queue still busy\n");
/* Wait for a bit */
init_waitqueue_head(&waitq);
lwi = LWI_TIMEOUT(cfs_time_seconds(2), NULL, NULL);
l_wait_event(waitq, 0, &lwi);
break;
}
}
/* notreached */
}
void
cfs_wi_shutdown (void)
{
int i;
if (cfs_wi_data.wi_scheds == NULL)
return;
for (i = 0; i < cfs_wi_data.wi_nsched; i++)
cfs_wi_sched_shutdown(&cfs_wi_data.wi_scheds[i]);
#ifdef __KERNEL__
cfs_spin_lock(&cfs_wi_data.wi_glock);
i = 2;
while (cfs_wi_data.wi_nthreads != 0) {
CDEBUG(IS_PO2(++i) ? D_WARNING : D_NET,
"waiting for %d threads to terminate\n",
cfs_wi_data.wi_nthreads);
cfs_spin_unlock(&cfs_wi_data.wi_glock);
cfs_pause(cfs_time_seconds(1));
cfs_spin_lock(&cfs_wi_data.wi_glock);
}
cfs_spin_unlock(&cfs_wi_data.wi_glock);
#endif
LIBCFS_FREE(cfs_wi_data.wi_scheds,
cfs_wi_data.wi_nsched * sizeof(cfs_wi_sched_t));
return;
}
int LL_PROC_PROTO(proc_console_min_delay_cs)
{
int rc, min_delay_cs;
cfs_sysctl_table_t dummy = *table;
cfs_duration_t d;
dummy.data = &min_delay_cs;
dummy.proc_handler = &proc_dointvec;
if (!write) { /* read */
min_delay_cs = cfs_duration_sec(libcfs_console_min_delay * 100);
rc = ll_proc_dointvec(&dummy, write, filp, buffer, lenp, ppos);
return rc;
}
/* write */
min_delay_cs = 0;
rc = ll_proc_dointvec(&dummy, write, filp, buffer, lenp, ppos);
if (rc < 0)
return rc;
if (min_delay_cs <= 0)
return -EINVAL;
d = cfs_time_seconds(min_delay_cs) / 100;
if (d == 0 || d > libcfs_console_max_delay)
return -EINVAL;
libcfs_console_min_delay = d;
return rc;
}
static int osd_scrub_next(const struct lu_env *env, struct osd_device *dev,
struct lu_fid *fid, uint64_t *oid)
{
struct l_wait_info lwi = { 0 };
struct lustre_scrub *scrub = &dev->od_scrub;
struct ptlrpc_thread *thread = &scrub->os_thread;
struct osd_otable_it *it = dev->od_otable_it;
struct lustre_mdt_attrs *lma = NULL;
nvlist_t *nvbuf = NULL;
int size = 0;
int rc = 0;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
if (likely(lwi.lwi_timeout > 0)) {
l_wait_event(thread->t_ctl_waitq,
!list_empty(&scrub->os_inconsistent_items) ||
!thread_is_running(thread),
&lwi);
if (unlikely(!thread_is_running(thread)))
RETURN(SCRUB_NEXT_EXIT);
}
}
if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
spin_lock(&scrub->os_lock);
thread_set_flags(thread, SVC_STOPPING);
spin_unlock(&scrub->os_lock);
RETURN(SCRUB_NEXT_CRASH);
}
if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
RETURN(SCRUB_NEXT_FATAL);
again:
if (nvbuf) {
nvlist_free(nvbuf);
nvbuf = NULL;
lma = NULL;
}
if (!list_empty(&scrub->os_inconsistent_items)) {
spin_lock(&scrub->os_lock);
if (likely(!list_empty(&scrub->os_inconsistent_items))) {
struct osd_inconsistent_item *oii;
oii = list_entry(scrub->os_inconsistent_items.next,
struct osd_inconsistent_item, oii_list);
*fid = oii->oii_cache.oic_fid;
*oid = oii->oii_cache.oic_dnode;
scrub->os_in_prior = 1;
spin_unlock(&scrub->os_lock);
GOTO(out, rc = 0);
}
spin_unlock(&scrub->os_lock);
}
/* Read from wire as much data as possible.
* Returns 0 or 1 on succsess, <0 if error or EOF.
* 0 means partial read, 1 - complete */
int
usocklnd_read_data(usock_conn_t *conn)
{
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (conn->uc_rx_nob_wanted != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (conn->uc_rx_niov > 0);
nob = libcfs_sock_readv(conn->uc_sock,
conn->uc_rx_iov, conn->uc_rx_niov);
if (nob <= 0) {/* read nothing or error */
if (nob < 0)
conn->uc_errored = 1;
return nob;
}
LASSERT (nob <= conn->uc_rx_nob_wanted);
conn->uc_rx_nob_wanted -= nob;
conn->uc_rx_nob_left -= nob;
t = cfs_time_current();
conn->uc_rx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = conn->uc_rx_iov;
do {
LASSERT (conn->uc_rx_niov > 0);
if (nob < iov->iov_len) {
iov->iov_base = (void *)(((unsigned long)(iov->iov_base)) + nob);
iov->iov_len -= nob;
break;
}
nob -= iov->iov_len;
conn->uc_rx_iov = ++iov;
conn->uc_rx_niov--;
} while (nob != 0);
} while (conn->uc_rx_nob_wanted != 0);
return 1; /* read complete */
}
int __cfs_fail_timeout_set(__u32 id, __u32 value, int ms, int set)
{
int ret;
ret = __cfs_fail_check_set(id, value, set);
if (ret && likely(ms > 0)) {
CERROR("cfs_fail_timeout id %x sleeping for %dms\n",
id, ms);
set_current_state(TASK_UNINTERRUPTIBLE);
schedule_timeout(cfs_time_seconds(ms) / 1000);
CERROR("cfs_fail_timeout id %x awake\n", id);
}
return ret;
}
/* Send as much tx data as possible.
* Returns 0 or 1 on succsess, <0 if fatal error.
* 0 means partial send or non-fatal error, 1 - complete.
* Rely on libcfs_sock_writev() for differentiating fatal and
* non-fatal errors. An error should be considered as non-fatal if:
* 1) it still makes sense to continue reading &&
* 2) anyway, poll() will set up POLLHUP|POLLERR flags */
int
usocklnd_send_tx(usock_conn_t *conn, usock_tx_t *tx)
{
struct iovec *iov;
int nob;
cfs_time_t t;
LASSERT (tx->tx_resid != 0);
do {
usock_peer_t *peer = conn->uc_peer;
LASSERT (tx->tx_niov > 0);
nob = libcfs_sock_writev(conn->uc_sock,
tx->tx_iov, tx->tx_niov);
if (nob < 0)
conn->uc_errored = 1;
if (nob <= 0) /* write queue is flow-controlled or error */
return nob;
LASSERT (nob <= tx->tx_resid);
tx->tx_resid -= nob;
t = cfs_time_current();
conn->uc_tx_deadline = cfs_time_add(t, cfs_time_seconds(usock_tuns.ut_timeout));
if(peer != NULL)
peer->up_last_alive = t;
/* "consume" iov */
iov = tx->tx_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;
break;
}
nob -= iov->iov_len;
tx->tx_iov = ++iov;
tx->tx_niov--;
} while (nob != 0);
} while (tx->tx_resid != 0);
return 1; /* send complete */
}
void lc_watchdog_touch(struct lc_watchdog *lcw, int timeout)
{
ENTRY;
LASSERT(lcw != NULL);
lc_watchdog_del_pending(lcw);
lcw_update_time(lcw, "resumed");
lcw->lcw_state = LC_WATCHDOG_ENABLED;
cfs_timer_arm(&lcw->lcw_timer, cfs_time_current() +
cfs_time_seconds(timeout));
EXIT;
}
int __obd_fail_timeout_set(__u32 id, __u32 value, int ms, int set)
{
int ret = 0;
ret = __obd_fail_check_set(id, value, set);
if (ret) {
CERROR("obd_fail_timeout id %x sleeping for %dms\n",
id, ms);
cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT,
cfs_time_seconds(ms) / 1000);
cfs_set_current_state(CFS_TASK_RUNNING);
CERROR("obd_fail_timeout id %x awake\n", id);
}
return ret;
}
static int param_set_delay_minmax(const char *val,
const struct kernel_param *kp,
long min, long max)
{
long d;
int sec;
int rc;
rc = kstrtoint(val, 0, &sec);
if (rc)
return -EINVAL;
d = cfs_time_seconds(sec) / 100;
if (d < min || d > max)
return -EINVAL;
*((unsigned int *)kp->arg) = d;
return 0;
}
struct lc_watchdog *lc_watchdog_add(int timeout,
void (*callback)(pid_t, void *),
void *data)
{
struct lc_watchdog *lcw = NULL;
ENTRY;
LIBCFS_ALLOC(lcw, sizeof(*lcw));
if (lcw == NULL) {
CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
RETURN(ERR_PTR(-ENOMEM));
}
cfs_spin_lock_init(&lcw->lcw_lock);
lcw->lcw_refcount = 1; /* refcount for owner */
lcw->lcw_task = cfs_current();
lcw->lcw_pid = cfs_curproc_pid();
lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
lcw->lcw_data = data;
lcw->lcw_state = LC_WATCHDOG_DISABLED;
CFS_INIT_LIST_HEAD(&lcw->lcw_list);
cfs_timer_init(&lcw->lcw_timer, lcw_cb, lcw);
cfs_down(&lcw_refcount_sem);
if (++lcw_refcount == 1)
lcw_dispatch_start();
cfs_up(&lcw_refcount_sem);
/* Keep this working in case we enable them by default */
if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
lcw->lcw_last_touched = cfs_time_current();
cfs_timer_arm(&lcw->lcw_timer, cfs_time_seconds(timeout) +
cfs_time_current());
}
RETURN(lcw);
}
/* Request sequence-controller node to allocate new meta-sequence. */
static int seq_client_alloc_meta(const struct lu_env *env,
struct lu_client_seq *seq)
{
int rc;
ENTRY;
if (seq->lcs_srv) {
#ifdef HAVE_SEQ_SERVER
LASSERT(env != NULL);
rc = seq_server_alloc_meta(seq->lcs_srv, &seq->lcs_space, env);
#else
rc = 0;
#endif
} else {
do {
/* If meta server return -EINPROGRESS or EAGAIN,
* it means meta server might not be ready to
* allocate super sequence from sequence controller
* (MDT0)yet */
rc = seq_client_rpc(seq, &seq->lcs_space,
SEQ_ALLOC_META, "meta");
if (rc == -EINPROGRESS || rc == -EAGAIN) {
wait_queue_head_t waitq;
struct l_wait_info lwi;
/* MDT0 is not ready, let's wait for 2
* seconds and retry. */
init_waitqueue_head(&waitq);
lwi = LWI_TIMEOUT(cfs_time_seconds(2), NULL,
NULL);
l_wait_event(waitq, 0, &lwi);
}
} while (rc == -EINPROGRESS || rc == -EAGAIN);
}
RETURN(rc);
}
/**
* Callback handler for receiving incoming completion ASTs.
*
* This only can happen on client side.
*/
static void ldlm_handle_cp_callback(struct ptlrpc_request *req,
struct ldlm_namespace *ns,
struct ldlm_request *dlm_req,
struct ldlm_lock *lock)
{
int lvb_len;
LIST_HEAD(ast_list);
int rc = 0;
LDLM_DEBUG(lock, "client completion callback handler START");
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
int to = cfs_time_seconds(1);
while (to > 0) {
schedule_timeout_and_set_state(
TASK_INTERRUPTIBLE, to);
if (lock->l_granted_mode == lock->l_req_mode ||
lock->l_flags & LDLM_FL_DESTROYED)
break;
}
}
lvb_len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB, RCL_CLIENT);
if (lvb_len < 0) {
LDLM_ERROR(lock, "Fail to get lvb_len, rc = %d", lvb_len);
GOTO(out, rc = lvb_len);
} else if (lvb_len > 0) {
if (lock->l_lvb_len > 0) {
/* for extent lock, lvb contains ost_lvb{}. */
LASSERT(lock->l_lvb_data != NULL);
if (unlikely(lock->l_lvb_len < lvb_len)) {
LDLM_ERROR(lock, "Replied LVB is larger than "
"expectation, expected = %d, "
"replied = %d",
lock->l_lvb_len, lvb_len);
GOTO(out, rc = -EINVAL);
}
} else if (ldlm_has_layout(lock)) { /* for layout lock, lvb has
* variable length */
void *lvb_data;
OBD_ALLOC(lvb_data, lvb_len);
if (lvb_data == NULL) {
LDLM_ERROR(lock, "No memory: %d.\n", lvb_len);
GOTO(out, rc = -ENOMEM);
}
lock_res_and_lock(lock);
LASSERT(lock->l_lvb_data == NULL);
lock->l_lvb_data = lvb_data;
lock->l_lvb_len = lvb_len;
unlock_res_and_lock(lock);
}
}
lock_res_and_lock(lock);
if ((lock->l_flags & LDLM_FL_DESTROYED) ||
lock->l_granted_mode == lock->l_req_mode) {
/* bug 11300: the lock has already been granted */
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "Double grant race happened");
GOTO(out, rc = 0);
}
/* If we receive the completion AST before the actual enqueue returned,
* then we might need to switch lock modes, resources, or extents. */
if (dlm_req->lock_desc.l_granted_mode != lock->l_req_mode) {
lock->l_req_mode = dlm_req->lock_desc.l_granted_mode;
LDLM_DEBUG(lock, "completion AST, new lock mode");
}
if (lock->l_resource->lr_type != LDLM_PLAIN) {
ldlm_convert_policy_to_local(req->rq_export,
dlm_req->lock_desc.l_resource.lr_type,
&dlm_req->lock_desc.l_policy_data,
&lock->l_policy_data);
LDLM_DEBUG(lock, "completion AST, new policy data");
}
ldlm_resource_unlink_lock(lock);
if (memcmp(&dlm_req->lock_desc.l_resource.lr_name,
&lock->l_resource->lr_name,
sizeof(lock->l_resource->lr_name)) != 0) {
unlock_res_and_lock(lock);
rc = ldlm_lock_change_resource(ns, lock,
&dlm_req->lock_desc.l_resource.lr_name);
if (rc < 0) {
LDLM_ERROR(lock, "Failed to allocate resource");
GOTO(out, rc);
}
LDLM_DEBUG(lock, "completion AST, new resource");
CERROR("change resource!\n");
lock_res_and_lock(lock);
}
if (dlm_req->lock_flags & LDLM_FL_AST_SENT) {
/* BL_AST locks are not needed in LRU.
* Let ldlm_cancel_lru() be fast. */
ldlm_lock_remove_from_lru(lock);
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST;
LDLM_DEBUG(lock, "completion AST includes blocking AST");
}
if (lock->l_lvb_len > 0) {
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_CLIENT,
lock->l_lvb_data, lvb_len);
if (rc < 0) {
unlock_res_and_lock(lock);
GOTO(out, rc);
}
}
ldlm_grant_lock(lock, &ast_list);
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "callback handler finished, about to run_ast_work");
/* Let Enqueue to call osc_lock_upcall() and initialize
* l_ast_data */
OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 2);
ldlm_run_ast_work(ns, &ast_list, LDLM_WORK_CP_AST);
LDLM_DEBUG_NOLOCK("client completion callback handler END (lock %p)",
lock);
GOTO(out, rc);
out:
if (rc < 0) {
lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_FAILED;
unlock_res_and_lock(lock);
wake_up(&lock->l_waitq);
}
LDLM_LOCK_RELEASE(lock);
}
/*
* Release qsd_qtype_info structure which contains data associated with a
* given quota type. This releases the accounting objects.
* It's called on OSD cleanup when the qsd instance is released.
*
* \param env - is the environment passed by the caller
* \param qsd - is the qsd instance managing the qsd_qtype_info structure
* to be released
* \param qtype - is the quota type to be shutdown
*/
static void qsd_qtype_fini(const struct lu_env *env, struct qsd_instance *qsd,
int qtype)
{
struct qsd_qtype_info *qqi;
int repeat = 0;
ENTRY;
if (qsd->qsd_type_array[qtype] == NULL)
RETURN_EXIT;
qqi = qsd->qsd_type_array[qtype];
qsd->qsd_type_array[qtype] = NULL;
/* all deferred work lists should be empty */
LASSERT(cfs_list_empty(&qqi->qqi_deferred_glb));
LASSERT(cfs_list_empty(&qqi->qqi_deferred_slv));
/* shutdown lquota site */
if (qqi->qqi_site != NULL && !IS_ERR(qqi->qqi_site)) {
lquota_site_free(env, qqi->qqi_site);
qqi->qqi_site = NULL;
}
/* The qqi may still be holding by global locks which are being
* canceled asynchronously (LU-4365), see the following steps:
*
* - On server umount, we try to clear all quota locks first by
* disconnecting LWP (which will invalidate import and cleanup
* all locks on it), however, if quota reint process is holding
* the global lock for reintegration at that time, global lock
* will fail to be cleared on LWP disconnection.
*
* - Umount process goes on and stops reint process, the global
* lock will be dropped on reint process exit, however, the lock
* cancel in done in asynchronous way, so the
* qsd_glb_blocking_ast() might haven't been called yet when we
* get here.
*/
while (atomic_read(&qqi->qqi_ref) > 1) {
CDEBUG(D_QUOTA, "qqi reference count %u, repeat: %d\n",
atomic_read(&qqi->qqi_ref), repeat);
repeat++;
schedule_timeout_and_set_state(TASK_INTERRUPTIBLE,
cfs_time_seconds(1));
}
/* by now, all qqi users should have gone away */
LASSERT(atomic_read(&qqi->qqi_ref) == 1);
lu_ref_fini(&qqi->qqi_reference);
/* release accounting object */
if (qqi->qqi_acct_obj != NULL && !IS_ERR(qqi->qqi_acct_obj)) {
lu_object_put(env, &qqi->qqi_acct_obj->do_lu);
qqi->qqi_acct_obj = NULL;
}
/* release slv index */
if (qqi->qqi_slv_obj != NULL && !IS_ERR(qqi->qqi_slv_obj)) {
lu_object_put(env, &qqi->qqi_slv_obj->do_lu);
qqi->qqi_slv_obj = NULL;
qqi->qqi_slv_ver = 0;
}
/* release global index */
if (qqi->qqi_glb_obj != NULL && !IS_ERR(qqi->qqi_glb_obj)) {
lu_object_put(env, &qqi->qqi_glb_obj->do_lu);
qqi->qqi_glb_obj = NULL;
qqi->qqi_glb_ver = 0;
}
OBD_FREE_PTR(qqi);
EXIT;
}
static int lfsck_master_oit_engine(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
const struct dt_it_ops *iops =
&lfsck->li_obj_oit->do_index_ops->dio_it;
struct dt_it *di = lfsck->li_di_oit;
struct lu_fid *fid = &info->lti_fid;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct ptlrpc_thread *thread = &lfsck->li_thread;
__u32 idx =
lfsck_dev_idx(lfsck->li_bottom);
int rc;
ENTRY;
do {
struct dt_object *target;
bool update_lma = false;
if (lfsck->li_di_dir != NULL) {
rc = lfsck_master_dir_engine(env, lfsck);
if (rc <= 0)
RETURN(rc);
}
if (unlikely(lfsck->li_oit_over))
RETURN(1);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
cfs_fail_val > 0) {
struct l_wait_info lwi;
lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
NULL, NULL);
l_wait_event(thread->t_ctl_waitq,
!thread_is_running(thread),
&lwi);
}
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
RETURN(0);
lfsck->li_current_oit_processed = 1;
lfsck->li_new_scanned++;
rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
if (rc != 0) {
lfsck_fail(env, lfsck, true);
if (rc < 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
else
goto checkpoint;
}
if (fid_is_idif(fid)) {
__u32 idx1 = fid_idif_ost_idx(fid);
LASSERT(!lfsck->li_master);
/* It is an old format device, update the LMA. */
if (idx != idx1) {
struct ost_id *oi = &info->lti_oi;
fid_to_ostid(fid, oi);
ostid_to_fid(fid, oi, idx);
update_lma = true;
}
} else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
!fid_is_last_id(fid) && !fid_is_root(fid) &&
!fid_seq_is_dot(fid_seq(fid))) {
/* If the FID/object is only used locally and invisible
* to external nodes, then LFSCK will not handle it. */
goto checkpoint;
}
target = lfsck_object_find(env, lfsck, fid);
if (target == NULL) {
goto checkpoint;
} else if (IS_ERR(target)) {
lfsck_fail(env, lfsck, true);
if (bk->lb_param & LPF_FAILOUT)
RETURN(PTR_ERR(target));
else
goto checkpoint;
}
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_exists(target) && !dt_object_remote(target)) {
if (update_lma)
rc = lfsck_update_lma(env, lfsck, target);
if (rc == 0)
rc = lfsck_exec_oit(env, lfsck, target);
}
lfsck_object_put(env, target);
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
checkpoint:
rc = lfsck_checkpoint(env, lfsck);
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
/* Rate control. */
lfsck_control_speed(lfsck);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
spin_lock(&lfsck->li_lock);
thread_set_flags(thread, SVC_STOPPING);
spin_unlock(&lfsck->li_lock);
RETURN(-EINVAL);
}
rc = iops->next(env, di);
if (unlikely(rc > 0))
lfsck->li_oit_over = 1;
else if (likely(rc == 0))
lfsck->li_current_oit_processed = 0;
if (unlikely(!thread_is_running(thread)))
RETURN(0);
} while (rc == 0 || lfsck->li_di_dir != NULL);
RETURN(rc);
}
int libcfs_debug_vmsg2(cfs_debug_limit_state_t *cdls, int subsys, int mask,
const char *file, const char *fn, const int line,
const char *format1, va_list args,
const char *format2, ...)
{
struct cfs_trace_cpu_data *tcd = NULL;
struct ptldebug_header header = {0};
struct cfs_trace_page *tage;
/* string_buf is used only if tcd != NULL, and is always set then */
char *string_buf = NULL;
char *debug_buf;
int known_size;
int needed = 85; /* average message length */
int max_nob;
va_list ap;
int depth;
int i;
int remain;
if (strchr(file, '/'))
file = strrchr(file, '/') + 1;
tcd = cfs_trace_get_tcd();
/* cfs_trace_get_tcd() grabs a lock, which disables preemption and
* pins us to a particular CPU. This avoids an smp_processor_id()
* warning on Linux when debugging is enabled. */
cfs_set_ptldebug_header(&header, subsys, mask, line, CDEBUG_STACK());
if (tcd == NULL) /* arch may not log in IRQ context */
goto console;
if (tcd->tcd_cur_pages == 0)
header.ph_flags |= PH_FLAG_FIRST_RECORD;
if (tcd->tcd_shutting_down) {
cfs_trace_put_tcd(tcd);
tcd = NULL;
goto console;
}
depth = __current_nesting_level();
known_size = strlen(file) + 1 + depth;
if (fn)
known_size += strlen(fn) + 1;
if (libcfs_debug_binary)
known_size += sizeof(header);
/*/
* '2' used because vsnprintf return real size required for output
* _without_ terminating NULL.
* if needed is to small for this format.
*/
for (i = 0; i < 2; i++) {
tage = cfs_trace_get_tage(tcd, needed + known_size + 1);
if (tage == NULL) {
if (needed + known_size > CFS_PAGE_SIZE)
mask |= D_ERROR;
cfs_trace_put_tcd(tcd);
tcd = NULL;
goto console;
}
string_buf = (char *)cfs_page_address(tage->page) +
tage->used + known_size;
max_nob = CFS_PAGE_SIZE - tage->used - known_size;
if (max_nob <= 0) {
printk(CFS_KERN_EMERG "negative max_nob: %d\n",
max_nob);
mask |= D_ERROR;
cfs_trace_put_tcd(tcd);
tcd = NULL;
goto console;
}
needed = 0;
if (format1) {
va_copy(ap, args);
needed = vsnprintf(string_buf, max_nob, format1, ap);
va_end(ap);
}
if (format2) {
remain = max_nob - needed;
if (remain < 0)
remain = 0;
va_start(ap, format2);
needed += vsnprintf(string_buf + needed, remain,
format2, ap);
va_end(ap);
}
if (needed < max_nob) /* well. printing ok.. */
break;
}
if (*(string_buf+needed-1) != '\n')
printk(CFS_KERN_INFO "format at %s:%d:%s doesn't end in newline\n",
file, line, fn);
header.ph_len = known_size + needed;
debug_buf = (char *)cfs_page_address(tage->page) + tage->used;
if (libcfs_debug_binary) {
memcpy(debug_buf, &header, sizeof(header));
tage->used += sizeof(header);
debug_buf += sizeof(header);
}
/* indent message according to the nesting level */
while (depth-- > 0) {
*(debug_buf++) = '.';
++ tage->used;
}
strcpy(debug_buf, file);
tage->used += strlen(file) + 1;
debug_buf += strlen(file) + 1;
if (fn) {
strcpy(debug_buf, fn);
tage->used += strlen(fn) + 1;
debug_buf += strlen(fn) + 1;
}
__LASSERT(debug_buf == string_buf);
tage->used += needed;
__LASSERT (tage->used <= CFS_PAGE_SIZE);
console:
if ((mask & libcfs_printk) == 0) {
/* no console output requested */
if (tcd != NULL)
cfs_trace_put_tcd(tcd);
return 1;
}
if (cdls != NULL) {
if (libcfs_console_ratelimit &&
cdls->cdls_next != 0 && /* not first time ever */
!cfs_time_after(cfs_time_current(), cdls->cdls_next)) {
/* skipping a console message */
cdls->cdls_count++;
if (tcd != NULL)
cfs_trace_put_tcd(tcd);
return 1;
}
if (cfs_time_after(cfs_time_current(), cdls->cdls_next +
libcfs_console_max_delay
+ cfs_time_seconds(10))) {
/* last timeout was a long time ago */
cdls->cdls_delay /= libcfs_console_backoff * 4;
} else {
cdls->cdls_delay *= libcfs_console_backoff;
if (cdls->cdls_delay < libcfs_console_min_delay)
cdls->cdls_delay = libcfs_console_min_delay;
else if (cdls->cdls_delay > libcfs_console_max_delay)
cdls->cdls_delay = libcfs_console_max_delay;
}
/* ensure cdls_next is never zero after it's been seen */
cdls->cdls_next = (cfs_time_current() + cdls->cdls_delay) | 1;
}
if (tcd != NULL) {
cfs_print_to_console(&header, mask, string_buf, needed, file,
fn);
cfs_trace_put_tcd(tcd);
} else {
string_buf = cfs_trace_get_console_buffer();
needed = 0;
if (format1 != NULL) {
va_copy(ap, args);
needed = vsnprintf(string_buf,
CFS_TRACE_CONSOLE_BUFFER_SIZE,
format1, ap);
va_end(ap);
}
if (format2 != NULL) {
remain = CFS_TRACE_CONSOLE_BUFFER_SIZE - needed;
if (remain > 0) {
va_start(ap, format2);
needed += vsnprintf(string_buf+needed, remain, format2, ap);
va_end(ap);
}
}
cfs_print_to_console(&header, mask,
string_buf, needed, file, fn);
cfs_trace_put_console_buffer(string_buf);
}
if (cdls != NULL && cdls->cdls_count != 0) {
string_buf = cfs_trace_get_console_buffer();
needed = snprintf(string_buf, CFS_TRACE_CONSOLE_BUFFER_SIZE,
"Skipped %d previous similar message%s\n",
cdls->cdls_count, (cdls->cdls_count > 1) ? "s" : "");
cfs_print_to_console(&header, mask,
string_buf, needed, file, fn);
cfs_trace_put_console_buffer(string_buf);
cdls->cdls_count = 0;
}
return 0;
}
int
libcfs_sock_read (cfs_socket_t *sock, void *buffer, int nob, int timeout)
{
size_t rcvlen;
int rc;
cfs_duration_t to = cfs_time_seconds(timeout);
cfs_time_t then;
struct timeval tv;
LASSERT(nob > 0);
for (;;) {
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,
};
cfs_duration_usec(to, &tv);
rc = -sock_setsockopt(C2B_SOCK(sock), SOL_SOCKET, SO_RCVTIMEO,
&tv, sizeof(tv));
if (rc != 0) {
CERROR("Can't set socket recv timeout "
"%ld.%06d: %d\n",
(long)tv.tv_sec, (int)tv.tv_usec, rc);
return rc;
}
then = cfs_time_current();
rc = -sock_receive(C2B_SOCK(sock), &msg, 0, &rcvlen);
to -= cfs_time_current() - then;
if (rc != 0 && rc != -EWOULDBLOCK)
return rc;
if (rcvlen == nob)
return 0;
if (to <= 0)
return -EAGAIN;
buffer = ((char *)buffer) + rcvlen;
nob -= rcvlen;
}
return 0;
}
int
libcfs_sock_write (cfs_socket_t *sock, void *buffer, int nob, int timeout)
{
size_t sndlen;
int rc;
cfs_duration_t to = cfs_time_seconds(timeout);
cfs_time_t then;
struct timeval tv;
LASSERT(nob > 0);
for (;;) {
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 = (timeout == 0) ? MSG_DONTWAIT : 0,
};
if (timeout != 0) {
cfs_duration_usec(to, &tv);
rc = -sock_setsockopt(C2B_SOCK(sock), SOL_SOCKET, SO_SNDTIMEO,
&tv, sizeof(tv));
if (rc != 0) {
CERROR("Can't set socket send timeout "
"%ld.%06d: %d\n",
(long)tv.tv_sec, (int)tv.tv_usec, rc);
return rc;
}
}
then = cfs_time_current();
rc = -sock_send(C2B_SOCK(sock), &msg,
((timeout == 0) ? MSG_DONTWAIT : 0), &sndlen);
to -= cfs_time_current() - then;
if (rc != 0 && rc != -EWOULDBLOCK)
return rc;
if (sndlen == nob)
return 0;
if (to <= 0)
return -EAGAIN;
buffer = ((char *)buffer) + sndlen;
nob -= sndlen;
}
return 0;
}
int
libcfs_sock_getaddr (cfs_socket_t *sock, int remote, __u32 *ip, int *port)
{
struct sockaddr_in sin;
int rc;
if (remote != 0)
/* Get remote address */
rc = -sock_getpeername(C2B_SOCK(sock), (struct sockaddr *)&sin, sizeof(sin));
else
/* Get local address */
rc = -sock_getsockname(C2B_SOCK(sock), (struct sockaddr *)&sin, sizeof(sin));
if (rc != 0) {
CERROR ("Error %d getting sock %s IP/port\n",
rc, remote ? "peer" : "local");
return rc;
}
if (ip != NULL)
*ip = ntohl (sin.sin_addr.s_addr);
if (port != NULL)
*port = ntohs (sin.sin_port);
return 0;
}
int
libcfs_sock_setbuf (cfs_socket_t *sock, int txbufsize, int rxbufsize)
{
int option;
int rc;
if (txbufsize != 0) {
option = txbufsize;
rc = -sock_setsockopt(C2B_SOCK(sock), SOL_SOCKET, SO_SNDBUF,
(char *)&option, sizeof (option));
if (rc != 0) {
CERROR ("Can't set send buffer %d: %d\n",
option, rc);
return (rc);
}
}
if (rxbufsize != 0) {
option = rxbufsize;
rc = -sock_setsockopt (C2B_SOCK(sock), SOL_SOCKET, SO_RCVBUF,
(char *)&option, sizeof (option));
if (rc != 0) {
CERROR ("Can't set receive buffer %d: %d\n",
option, rc);
return (rc);
}
}
return 0;
}
static inline cfs_time_t capa_renewal_time(struct obd_capa *ocapa)
{
return cfs_time_sub(ocapa->c_expiry,
cfs_time_seconds(ocapa->c_capa.lc_timeout) / 2);
}
int
ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx)
{
#if SOCKNAL_SINGLE_FRAG_TX
struct iovec scratch;
struct iovec *scratchiov = &scratch;
unsigned int niov = 1;
#else
struct iovec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
unsigned int niov = tx->tx_niov;
#endif
struct socket *sock = conn->ksnc_sock;
int nob;
int rc;
int i;
struct uio suio = {
.uio_iov = scratchiov,
.uio_iovcnt = niov,
.uio_offset = 0,
.uio_resid = 0, /* This will be valued after a while */
.uio_segflg = UIO_SYSSPACE,
.uio_rw = UIO_WRITE,
.uio_procp = NULL
};
int flags = MSG_DONTWAIT;
CFS_DECL_NET_DATA;
for (nob = i = 0; i < niov; i++) {
scratchiov[i] = tx->tx_iov[i];
nob += scratchiov[i].iov_len;
}
suio.uio_resid = nob;
CFS_NET_IN;
rc = sosend(sock, NULL, &suio, (struct mbuf *)0, (struct mbuf *)0, flags);
CFS_NET_EX;
/* NB there is no return value can indicate how many
* have been sent and how many resid, we have to get
* sent bytes from suio. */
if (rc != 0) {
if (suio.uio_resid != nob &&\
(rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
/* We have sent something */
rc = nob - suio.uio_resid;
else if ( rc == EWOULDBLOCK )
/* Actually, EAGAIN and EWOULDBLOCK have same value in OSX */
rc = -EAGAIN;
else
rc = -rc;
} else /* rc == 0 */
rc = nob - suio.uio_resid;
return rc;
}
int
ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx)
{
#if SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_RISK_KMAP_DEADLOCK
struct iovec scratch;
struct iovec *scratchiov = &scratch;
unsigned int niov = 1;
#else
struct iovec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
unsigned int niov = tx->tx_nkiov;
#endif
struct socket *sock = conn->ksnc_sock;
lnet_kiov_t *kiov = tx->tx_kiov;
int nob;
int rc;
int i;
struct uio suio = {
.uio_iov = scratchiov,
.uio_iovcnt = niov,
.uio_offset = 0,
.uio_resid = 0, /* It should be valued after a while */
.uio_segflg = UIO_SYSSPACE,
.uio_rw = UIO_WRITE,
.uio_procp = NULL
};
int flags = MSG_DONTWAIT;
CFS_DECL_NET_DATA;
for (nob = i = 0; i < niov; i++) {
scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) +
kiov[i].kiov_offset;
nob += scratchiov[i].iov_len = kiov[i].kiov_len;
}
suio.uio_resid = nob;
CFS_NET_IN;
rc = sosend(sock, NULL, &suio, (struct mbuf *)0, (struct mbuf *)0, flags);
CFS_NET_EX;
for (i = 0; i < niov; i++)
cfs_kunmap(kiov[i].kiov_page);
if (rc != 0) {
if (suio.uio_resid != nob &&\
(rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
/* We have sent something */
rc = nob - suio.uio_resid;
else if ( rc == EWOULDBLOCK )
/* EAGAIN and EWOULD BLOCK have same value in OSX */
rc = -EAGAIN;
else
rc = -rc;
} else /* rc == 0 */
rc = nob - suio.uio_resid;
return rc;
}
/*
* liang: Hack of inpcb and tcpcb.
* To get tcpcb of a socket, and call tcp_output
* to send quick ack.
*/
struct ks_tseg_qent{
int foo;
};
struct ks_tcptemp{
int foo;
};
LIST_HEAD(ks_tsegqe_head, ks_tseg_qent);
struct ks_tcpcb {
struct ks_tsegqe_head t_segq;
int t_dupacks;
struct ks_tcptemp *unused;
int t_timer[4];
struct inpcb *t_inpcb;
int t_state;
u_int t_flags;
/*
* There are more fields but we dont need
* ......
*/
};
#define TF_ACKNOW 0x00001
#define TF_DELACK 0x00002
struct ks_inpcb {
LIST_ENTRY(ks_inpcb) inp_hash;
struct in_addr reserved1;
struct in_addr reserved2;
u_short inp_fport;
u_short inp_lport;
LIST_ENTRY(inpcb) inp_list;
caddr_t inp_ppcb;
/*
* There are more fields but we dont need
* ......
*/
};
#define ks_sotoinpcb(so) ((struct ks_inpcb *)(so)->so_pcb)
#define ks_intotcpcb(ip) ((struct ks_tcpcb *)(ip)->inp_ppcb)
#define ks_sototcpcb(so) (intotcpcb(sotoinpcb(so)))
void
ksocknal_lib_eager_ack (ksock_conn_t *conn)
{
struct socket *sock = conn->ksnc_sock;
struct ks_inpcb *inp = ks_sotoinpcb(sock);
struct ks_tcpcb *tp = ks_intotcpcb(inp);
int s;
CFS_DECL_NET_DATA;
extern int tcp_output(register struct ks_tcpcb *tp);
CFS_NET_IN;
s = splnet();
/*
* No TCP_QUICKACK supported in BSD, so I have to call tcp_fasttimo
* to send immediate ACK.
*/
if (tp && tp->t_flags & TF_DELACK){
tp->t_flags &= ~TF_DELACK;
tp->t_flags |= TF_ACKNOW;
(void) tcp_output(tp);
}
splx(s);
CFS_NET_EX;
return;
}
int
ksocknal_lib_recv_iov (ksock_conn_t *conn)
{
#if SOCKNAL_SINGLE_FRAG_RX
struct iovec scratch;
struct iovec *scratchiov = &scratch;
unsigned int niov = 1;
#else
struct iovec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
unsigned int niov = conn->ksnc_rx_niov;
#endif
struct iovec *iov = conn->ksnc_rx_iov;
int nob;
int rc;
int i;
struct uio ruio = {
.uio_iov = scratchiov,
.uio_iovcnt = niov,
.uio_offset = 0,
.uio_resid = 0, /* It should be valued after a while */
.uio_segflg = UIO_SYSSPACE,
.uio_rw = UIO_READ,
.uio_procp = NULL
};
int flags = MSG_DONTWAIT;
CFS_DECL_NET_DATA;
for (nob = i = 0; i < niov; i++) {
scratchiov[i] = iov[i];
nob += scratchiov[i].iov_len;
}
LASSERT (nob <= conn->ksnc_rx_nob_wanted);
ruio.uio_resid = nob;
CFS_NET_IN;
rc = soreceive(conn->ksnc_sock, (struct sockaddr **)0, &ruio, (struct mbuf **)0, (struct mbuf **)0, &flags);
CFS_NET_EX;
if (rc){
if (ruio.uio_resid != nob && \
(rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK || rc == EAGAIN))
/* data particially received */
rc = nob - ruio.uio_resid;
else if (rc == EWOULDBLOCK)
/* EAGAIN and EWOULD BLOCK have same value in OSX */
rc = -EAGAIN;
else
rc = -rc;
} else
rc = nob - ruio.uio_resid;
return (rc);
}
int
ksocknal_lib_recv_kiov (ksock_conn_t *conn)
{
#if SOCKNAL_SINGLE_FRAG_RX || !SOCKNAL_RISK_KMAP_DEADLOCK
struct iovec scratch;
struct iovec *scratchiov = &scratch;
unsigned int niov = 1;
#else
struct iovec *scratchiov = conn->ksnc_scheduler->kss_scratch_iov;
unsigned int niov = conn->ksnc_rx_nkiov;
#endif
lnet_kiov_t *kiov = conn->ksnc_rx_kiov;
int nob;
int rc;
int i;
struct uio ruio = {
.uio_iov = scratchiov,
.uio_iovcnt = niov,
.uio_offset = 0,
.uio_resid = 0,
.uio_segflg = UIO_SYSSPACE,
.uio_rw = UIO_READ,
.uio_procp = NULL
};
int flags = MSG_DONTWAIT;
CFS_DECL_NET_DATA;
for (nob = i = 0; i < niov; i++) {
scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) + kiov[i].kiov_offset;
nob += scratchiov[i].iov_len = kiov[i].kiov_len;
}
LASSERT (nob <= conn->ksnc_rx_nob_wanted);
ruio.uio_resid = nob;
CFS_NET_IN;
rc = soreceive(conn->ksnc_sock, (struct sockaddr **)0, &ruio, (struct mbuf **)0, NULL, &flags);
CFS_NET_EX;
for (i = 0; i < niov; i++)
cfs_kunmap(kiov[i].kiov_page);
if (rc){
if (ruio.uio_resid != nob && \
(rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
/* data particially received */
rc = nob - ruio.uio_resid;
else if (rc == EWOULDBLOCK)
/* receive blocked, EWOULDBLOCK == EAGAIN */
rc = -EAGAIN;
else
rc = -rc;
} else
rc = nob - ruio.uio_resid;
return (rc);
}
int
ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
{
struct socket *sock = conn->ksnc_sock;
int rc;
rc = ksocknal_connsock_addref(conn);
if (rc != 0) {
LASSERT (conn->ksnc_closing);
*txmem = *rxmem = *nagle = 0;
return -ESHUTDOWN;
}
rc = libcfs_sock_getbuf(sock, txmem, rxmem);
if (rc == 0) {
struct sockopt sopt;
int len;
CFS_DECL_NET_DATA;
len = sizeof(*nagle);
bzero(&sopt, sizeof sopt);
sopt.sopt_dir = SOPT_GET;
sopt.sopt_level = IPPROTO_TCP;
sopt.sopt_name = TCP_NODELAY;
sopt.sopt_val = nagle;
sopt.sopt_valsize = len;
CFS_NET_IN;
rc = -sogetopt(sock, &sopt);
CFS_NET_EX;
}
ksocknal_connsock_decref(conn);
if (rc == 0)
*nagle = !*nagle;
else
*txmem = *rxmem = *nagle = 0;
return (rc);
}
int
ksocknal_lib_setup_sock (struct socket *so)
{
struct sockopt sopt;
int rc;
int option;
int keep_idle;
int keep_intvl;
int keep_count;
int do_keepalive;
struct linger linger;
CFS_DECL_NET_DATA;
rc = libcfs_sock_setbuf(so,
*ksocknal_tunables.ksnd_tx_buffer_size,
*ksocknal_tunables.ksnd_rx_buffer_size);
if (rc != 0) {
CERROR ("Can't set buffer tx %d, rx %d buffers: %d\n",
*ksocknal_tunables.ksnd_tx_buffer_size,
*ksocknal_tunables.ksnd_rx_buffer_size, rc);
return (rc);
}
/* Ensure this socket aborts active sends immediately when we close
* it. */
bzero(&sopt, sizeof sopt);
linger.l_onoff = 0;
linger.l_linger = 0;
sopt.sopt_dir = SOPT_SET;
sopt.sopt_level = SOL_SOCKET;
sopt.sopt_name = SO_LINGER;
sopt.sopt_val = &linger;
sopt.sopt_valsize = sizeof(linger);
CFS_NET_IN;
rc = -sosetopt(so, &sopt);
if (rc != 0) {
CERROR ("Can't set SO_LINGER: %d\n", rc);
goto out;
}
if (!*ksocknal_tunables.ksnd_nagle) {
option = 1;
bzero(&sopt, sizeof sopt);
sopt.sopt_dir = SOPT_SET;
sopt.sopt_level = IPPROTO_TCP;
sopt.sopt_name = TCP_NODELAY;
sopt.sopt_val = &option;
sopt.sopt_valsize = sizeof(option);
rc = -sosetopt(so, &sopt);
if (rc != 0) {
CERROR ("Can't disable nagle: %d\n", rc);
goto out;
}
}
/* 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);
bzero(&sopt, sizeof sopt);
sopt.sopt_dir = SOPT_SET;
sopt.sopt_level = SOL_SOCKET;
sopt.sopt_name = SO_KEEPALIVE;
sopt.sopt_val = &option;
sopt.sopt_valsize = sizeof(option);
rc = -sosetopt(so, &sopt);
if (rc != 0) {
CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
goto out;
}
if (!do_keepalive) {
/* no more setting, just return */
rc = 0;
goto out;
}
bzero(&sopt, sizeof sopt);
sopt.sopt_dir = SOPT_SET;
sopt.sopt_level = IPPROTO_TCP;
sopt.sopt_name = TCP_KEEPALIVE;
sopt.sopt_val = &keep_idle;
sopt.sopt_valsize = sizeof(keep_idle);
rc = -sosetopt(so, &sopt);
if (rc != 0) {
CERROR ("Can't set TCP_KEEPALIVE : %d\n", rc);
goto out;
}
out:
CFS_NET_EX;
return (rc);
}
void
ksocknal_lib_push_conn(ksock_conn_t *conn)
{
struct socket *sock;
struct sockopt sopt;
int val = 1;
int rc;
CFS_DECL_NET_DATA;
rc = ksocknal_connsock_addref(conn);
if (rc != 0) /* being shut down */
return;
sock = conn->ksnc_sock;
bzero(&sopt, sizeof sopt);
sopt.sopt_dir = SOPT_SET;
sopt.sopt_level = IPPROTO_TCP;
sopt.sopt_name = TCP_NODELAY;
sopt.sopt_val = &val;
sopt.sopt_valsize = sizeof val;
CFS_NET_IN;
sosetopt(sock, &sopt);
CFS_NET_EX;
ksocknal_connsock_decref(conn);
return;
}
extern void ksocknal_read_callback (ksock_conn_t *conn);
extern void ksocknal_write_callback (ksock_conn_t *conn);
static void
ksocknal_upcall(struct socket *so, caddr_t arg, int waitf)
{
ksock_conn_t *conn = (ksock_conn_t *)arg;
ENTRY;
read_lock (&ksocknal_data.ksnd_global_lock);
if (conn == NULL)
goto out;
if (so->so_rcv.sb_flags & SB_UPCALL) {
extern int soreadable(struct socket *so);
if (conn->ksnc_rx_nob_wanted && soreadable(so))
/* To verify whether the upcall is for receive */
ksocknal_read_callback (conn);
}
/* go foward? */
if (so->so_snd.sb_flags & SB_UPCALL){
extern int sowriteable(struct socket *so);
if (sowriteable(so))
/* socket is writable */
ksocknal_write_callback(conn);
}
out:
read_unlock (&ksocknal_data.ksnd_global_lock);
EXIT;
}
void
ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn)
{
/* No callback need to save in osx */
return;
}
void
ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn)
{
CFS_DECL_NET_DATA;
CFS_NET_IN;
sock->so_upcallarg = (void *)conn;
sock->so_upcall = ksocknal_upcall;
sock->so_snd.sb_timeo = 0;
sock->so_rcv.sb_timeo = cfs_time_seconds(2);
sock->so_rcv.sb_flags |= SB_UPCALL;
sock->so_snd.sb_flags |= SB_UPCALL;
CFS_NET_EX;
return;
}
void
ksocknal_lib_act_callback(struct socket *sock, ksock_conn_t *conn)
{
CFS_DECL_NET_DATA;
CFS_NET_IN;
ksocknal_upcall (sock, (void *)conn, 0);
CFS_NET_EX;
}
int
LNetEQPoll (lnet_handle_eq_t *eventqs, int neq, int timeout_ms,
lnet_event_t *event, int *which)
{
int i;
int rc;
#ifdef __KERNEL__
cfs_waitlink_t wl;
cfs_time_t now;
#else
struct timeval then;
struct timeval now;
# ifdef HAVE_LIBPTHREAD
struct timespec ts;
# endif
lnet_ni_t *eqwaitni = the_lnet.ln_eqwaitni;
#endif
ENTRY;
LASSERT (the_lnet.ln_init);
LASSERT (the_lnet.ln_refcount > 0);
if (neq < 1)
RETURN(-ENOENT);
LNET_LOCK();
for (;;) {
#ifndef __KERNEL__
LNET_UNLOCK();
/* Recursion breaker */
if (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING &&
!LNetHandleIsEqual(eventqs[0], the_lnet.ln_rc_eqh))
lnet_router_checker();
LNET_LOCK();
#endif
for (i = 0; i < neq; i++) {
lnet_eq_t *eq = lnet_handle2eq(&eventqs[i]);
if (eq == NULL) {
LNET_UNLOCK();
RETURN(-ENOENT);
}
rc = lib_get_event (eq, event);
if (rc != 0) {
LNET_UNLOCK();
*which = i;
RETURN(rc);
}
}
#ifdef __KERNEL__
if (timeout_ms == 0) {
LNET_UNLOCK();
RETURN (0);
}
cfs_waitlink_init(&wl);
set_current_state(TASK_INTERRUPTIBLE);
cfs_waitq_add(&the_lnet.ln_waitq, &wl);
LNET_UNLOCK();
if (timeout_ms < 0) {
cfs_waitq_wait (&wl, CFS_TASK_INTERRUPTIBLE);
} else {
struct timeval tv;
now = cfs_time_current();
cfs_waitq_timedwait(&wl, CFS_TASK_INTERRUPTIBLE,
cfs_time_seconds(timeout_ms)/1000);
cfs_duration_usec(cfs_time_sub(cfs_time_current(), now),
&tv);
timeout_ms -= tv.tv_sec * 1000 + tv.tv_usec / 1000;
if (timeout_ms < 0)
timeout_ms = 0;
}
LNET_LOCK();
cfs_waitq_del(&the_lnet.ln_waitq, &wl);
#else
if (eqwaitni != NULL) {
/* I have a single NI that I have to call into, to get
* events queued, or to block. */
lnet_ni_addref_locked(eqwaitni);
LNET_UNLOCK();
if (timeout_ms <= 0) {
(eqwaitni->ni_lnd->lnd_wait)(eqwaitni, timeout_ms);
} else {
gettimeofday(&then, NULL);
(eqwaitni->ni_lnd->lnd_wait)(eqwaitni, timeout_ms);
gettimeofday(&now, NULL);
timeout_ms -= (now.tv_sec - then.tv_sec) * 1000 +
(now.tv_usec - then.tv_usec) / 1000;
if (timeout_ms < 0)
timeout_ms = 0;
}
LNET_LOCK();
lnet_ni_decref_locked(eqwaitni);
/* don't call into eqwaitni again if timeout has
* expired */
if (timeout_ms == 0)
eqwaitni = NULL;
continue; /* go back and check for events */
}
if (timeout_ms == 0) {
LNET_UNLOCK();
RETURN (0);
}
# ifndef HAVE_LIBPTHREAD
/* If I'm single-threaded, LNET fails at startup if it can't
* set the_lnet.ln_eqwaitni correctly. */
LBUG();
# else
if (timeout_ms < 0) {
pthread_cond_wait(&the_lnet.ln_cond,
&the_lnet.ln_lock);
} else {
gettimeofday(&then, NULL);
ts.tv_sec = then.tv_sec + timeout_ms/1000;
ts.tv_nsec = then.tv_usec * 1000 +
(timeout_ms%1000) * 1000000;
if (ts.tv_nsec >= 1000000000) {
ts.tv_sec++;
ts.tv_nsec -= 1000000000;
}
pthread_cond_timedwait(&the_lnet.ln_cond,
&the_lnet.ln_lock, &ts);
gettimeofday(&now, NULL);
timeout_ms -= (now.tv_sec - then.tv_sec) * 1000 +
(now.tv_usec - then.tv_usec) / 1000;
if (timeout_ms < 0)
timeout_ms = 0;
}
# endif
#endif
}
}
static int quota_chk_acq_common(struct obd_device *obd, struct obd_export *exp,
const unsigned int id[], int pending[],
int count, quota_acquire acquire,
struct obd_trans_info *oti, int isblk,
struct inode *inode, int frags)
{
struct lustre_quota_ctxt *qctxt = &obd->u.obt.obt_qctxt;
struct timeval work_start;
struct timeval work_end;
long timediff;
struct l_wait_info lwi = { 0 };
int rc = 0, cycle = 0, count_err = 1;
ENTRY;
if (!quota_is_set(obd, id, isblk ? QB_SET : QI_SET))
RETURN(0);
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* If the client has been evicted or if it
* timed out and tried to reconnect already,
* abort the request immediately */
RETURN(-ENOTCONN);
CDEBUG(D_QUOTA, "check quota for %s\n", obd->obd_name);
pending[USRQUOTA] = pending[GRPQUOTA] = 0;
/* Unfortunately, if quota master is too busy to handle the
* pre-dqacq in time and quota hash on ost is used up, we
* have to wait for the completion of in flight dqacq/dqrel,
* in order to get enough quota for write b=12588 */
cfs_gettimeofday(&work_start);
while ((rc = quota_check_common(obd, id, pending, count, cycle, isblk,
inode, frags)) &
QUOTA_RET_ACQUOTA) {
cfs_spin_lock(&qctxt->lqc_lock);
if (!qctxt->lqc_import && oti) {
cfs_spin_unlock(&qctxt->lqc_lock);
LASSERT(oti && oti->oti_thread &&
oti->oti_thread->t_watchdog);
lc_watchdog_disable(oti->oti_thread->t_watchdog);
CDEBUG(D_QUOTA, "sleep for quota master\n");
l_wait_event(qctxt->lqc_wait_for_qmaster, check_qm(qctxt),
&lwi);
CDEBUG(D_QUOTA, "wake up when quota master is back\n");
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
} else {
cfs_spin_unlock(&qctxt->lqc_lock);
}
cycle++;
if (isblk)
OBD_FAIL_TIMEOUT(OBD_FAIL_OST_HOLD_WRITE_RPC, 90);
/* after acquire(), we should run quota_check_common again
* so that we confirm there are enough quota to finish write */
rc = acquire(obd, id, oti, isblk);
/* please reference to dqacq_completion for the below */
/* a new request is finished, try again */
if (rc == QUOTA_REQ_RETURNED) {
CDEBUG(D_QUOTA, "finish a quota req, try again\n");
continue;
}
/* it is out of quota already */
if (rc == -EDQUOT) {
CDEBUG(D_QUOTA, "out of quota, return -EDQUOT\n");
break;
}
/* Related quota has been disabled by master, but enabled by
* slave, do not try again. */
if (unlikely(rc == -ESRCH)) {
CERROR("mismatched quota configuration, stop try.\n");
break;
}
if (isblk && (exp->exp_failed || exp->exp_abort_active_req))
/* The client has been evicted or tried to
* to reconnect already, abort the request */
RETURN(-ENOTCONN);
/* -EBUSY and others, wait a second and try again */
if (rc < 0) {
cfs_waitq_t waitq;
struct l_wait_info lwi;
if (oti && oti->oti_thread && oti->oti_thread->t_watchdog)
lc_watchdog_touch(oti->oti_thread->t_watchdog,
CFS_GET_TIMEOUT(oti->oti_thread->t_svc));
CDEBUG(D_QUOTA, "rc: %d, count_err: %d\n", rc,
count_err++);
cfs_waitq_init(&waitq);
lwi = LWI_TIMEOUT(cfs_time_seconds(min(cycle, 10)), NULL,
NULL);
l_wait_event(waitq, 0, &lwi);
}
if (rc < 0 || cycle % 10 == 0) {
cfs_spin_lock(&last_print_lock);
if (last_print == 0 ||
cfs_time_before((last_print + cfs_time_seconds(30)),
cfs_time_current())) {
last_print = cfs_time_current();
cfs_spin_unlock(&last_print_lock);
CWARN("still haven't managed to acquire quota "
"space from the quota master after %d "
"retries (err=%d, rc=%d)\n",
cycle, count_err - 1, rc);
} else {
cfs_spin_unlock(&last_print_lock);
}
}
CDEBUG(D_QUOTA, "recheck quota with rc: %d, cycle: %d\n", rc,
cycle);
}
cfs_gettimeofday(&work_end);
timediff = cfs_timeval_sub(&work_end, &work_start, NULL);
lprocfs_counter_add(qctxt->lqc_stats,
isblk ? LQUOTA_WAIT_FOR_CHK_BLK :
LQUOTA_WAIT_FOR_CHK_INO,
timediff);
if (rc > 0)
rc = 0;
RETURN(rc);
}
int quota_adjust_slave_lqs(struct quota_adjust_qunit *oqaq,
struct lustre_quota_ctxt *qctxt)
{
struct lustre_qunit_size *lqs = NULL;
unsigned long *unit, *tune;
signed long tmp = 0;
cfs_time_t time_limit = 0, *shrink;
int i, rc = 0;
ENTRY;
LASSERT(qctxt);
lqs = quota_search_lqs(LQS_KEY(QAQ_IS_GRP(oqaq), oqaq->qaq_id),
qctxt, QAQ_IS_CREATE_LQS(oqaq) ? 1 : 0);
if (lqs == NULL || IS_ERR(lqs)){
CERROR("fail to find a lqs for %sid %u!\n",
QAQ_IS_GRP(oqaq) ? "g" : "u", oqaq->qaq_id);
RETURN(PTR_ERR(lqs));
}
CDEBUG(D_QUOTA, "before: bunit: %lu, iunit: %lu.\n",
lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
cfs_spin_lock(&lqs->lqs_lock);
for (i = 0; i < 2; i++) {
if (i == 0 && !QAQ_IS_ADJBLK(oqaq))
continue;
if (i == 1 && !QAQ_IS_ADJINO(oqaq))
continue;
tmp = i ? (lqs->lqs_iunit_sz - oqaq->qaq_iunit_sz) :
(lqs->lqs_bunit_sz - oqaq->qaq_bunit_sz);
shrink = i ? &lqs->lqs_last_ishrink :
&lqs->lqs_last_bshrink;
time_limit = cfs_time_add(i ? lqs->lqs_last_ishrink :
lqs->lqs_last_bshrink,
cfs_time_seconds(qctxt->lqc_switch_seconds));
unit = i ? &lqs->lqs_iunit_sz : &lqs->lqs_bunit_sz;
tune = i ? &lqs->lqs_itune_sz : &lqs->lqs_btune_sz;
/* quota master shrinks */
if (qctxt->lqc_handler && tmp > 0)
*shrink = cfs_time_current();
/* quota master enlarges */
if (qctxt->lqc_handler && tmp < 0) {
/* in case of ping-pong effect, don't enlarge lqs
* in a short time */
if (*shrink &&
cfs_time_before(cfs_time_current(), time_limit))
tmp = 0;
}
/* when setquota, don't enlarge lqs b=18616 */
if (QAQ_IS_CREATE_LQS(oqaq) && tmp < 0)
tmp = 0;
if (tmp != 0) {
*unit = i ? oqaq->qaq_iunit_sz : oqaq->qaq_bunit_sz;
*tune = (*unit) / 2;
}
if (tmp > 0)
rc |= i ? LQS_INO_DECREASE : LQS_BLK_DECREASE;
if (tmp < 0)
rc |= i ? LQS_INO_INCREASE : LQS_BLK_INCREASE;
}
cfs_spin_unlock(&lqs->lqs_lock);
CDEBUG(D_QUOTA, "after: bunit: %lu, iunit: %lu.\n",
lqs->lqs_bunit_sz, lqs->lqs_iunit_sz);
lqs_putref(lqs);
RETURN(rc);
}
inline cfs_time_t round_timeout(cfs_time_t timeout)
{
return cfs_time_seconds((int)cfs_duration_sec(cfs_time_sub(timeout, 0)) + 1);
}
int
lnet_acceptor(void *arg)
{
socket_t *newsock;
int rc;
__u32 magic;
__u32 peer_ip;
int peer_port;
int secure = (int)((long_ptr_t)arg);
LASSERT(lnet_acceptor_state.pta_sock == NULL);
cfs_block_allsigs();
rc = libcfs_sock_listen(&lnet_acceptor_state.pta_sock,
0, accept_port, accept_backlog);
if (rc != 0) {
if (rc == -EADDRINUSE)
LCONSOLE_ERROR_MSG(0x122, "Can't start acceptor on port %d: port already in use\n",
accept_port);
else
LCONSOLE_ERROR_MSG(0x123, "Can't start acceptor on port %d: unexpected error %d\n",
accept_port, rc);
lnet_acceptor_state.pta_sock = NULL;
} else {
LCONSOLE(0, "Accept %s, port %d\n", accept_type, accept_port);
}
/* set init status and unblock parent */
lnet_acceptor_state.pta_shutdown = rc;
complete(&lnet_acceptor_state.pta_signal);
if (rc != 0)
return rc;
while (!lnet_acceptor_state.pta_shutdown) {
rc = libcfs_sock_accept(&newsock, lnet_acceptor_state.pta_sock);
if (rc != 0) {
if (rc != -EAGAIN) {
CWARN("Accept error %d: pausing...\n", rc);
cfs_pause(cfs_time_seconds(1));
}
continue;
}
/* maybe we're waken up with libcfs_sock_abort_accept() */
if (lnet_acceptor_state.pta_shutdown) {
libcfs_sock_release(newsock);
break;
}
rc = libcfs_sock_getaddr(newsock, 1, &peer_ip, &peer_port);
if (rc != 0) {
CERROR("Can't determine new connection's address\n");
goto failed;
}
if (secure && peer_port > LNET_ACCEPTOR_MAX_RESERVED_PORT) {
CERROR("Refusing connection from %pI4h: insecure port %d\n",
&peer_ip, peer_port);
goto failed;
}
rc = libcfs_sock_read(newsock, &magic, sizeof(magic),
accept_timeout);
if (rc != 0) {
CERROR("Error %d reading connection request from %pI4h\n",
rc, &peer_ip);
goto failed;
}
rc = lnet_accept(newsock, magic);
if (rc != 0)
goto failed;
continue;
failed:
libcfs_sock_release(newsock);
}
libcfs_sock_release(lnet_acceptor_state.pta_sock);
lnet_acceptor_state.pta_sock = NULL;
CDEBUG(D_NET, "Acceptor stopping\n");
/* unblock lnet_acceptor_stop() */
complete(&lnet_acceptor_state.pta_signal);
return 0;
}