* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include "ptlrpc_internal.h"
+const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_pin_ops = {
+ .add_kiov_frag = ptlrpc_prep_bulk_page_pin,
+ .release_frags = ptlrpc_release_bulk_page_pin,
+};
+EXPORT_SYMBOL(ptlrpc_bulk_kiov_pin_ops);
+
+const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kiov_nopin_ops = {
+ .add_kiov_frag = ptlrpc_prep_bulk_page_nopin,
+ .release_frags = ptlrpc_release_bulk_noop,
+};
+EXPORT_SYMBOL(ptlrpc_bulk_kiov_nopin_ops);
+
+const struct ptlrpc_bulk_frag_ops ptlrpc_bulk_kvec_ops = {
+ .add_iov_frag = ptlrpc_prep_bulk_frag,
+};
+EXPORT_SYMBOL(ptlrpc_bulk_kvec_ops);
+
static int ptlrpc_send_new_req(struct ptlrpc_request *req);
static int ptlrpcd_check_work(struct ptlrpc_request *req);
static int ptlrpc_unregister_reply(struct ptlrpc_request *request, int async);
* Allocate and initialize new bulk descriptor on the sender.
* Returns pointer to the descriptor or NULL on error.
*/
-struct ptlrpc_bulk_desc *ptlrpc_new_bulk(unsigned npages, unsigned max_brw,
- unsigned type, unsigned portal)
+struct ptlrpc_bulk_desc *ptlrpc_new_bulk(unsigned nfrags, unsigned max_brw,
+ enum ptlrpc_bulk_op_type type,
+ unsigned portal,
+ const struct ptlrpc_bulk_frag_ops *ops)
{
struct ptlrpc_bulk_desc *desc;
int i;
- OBD_ALLOC(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[npages]));
+ /* ensure that only one of KIOV or IOVEC is set but not both */
+ LASSERT((ptlrpc_is_bulk_desc_kiov(type) &&
+ ops->add_kiov_frag != NULL) ||
+ (ptlrpc_is_bulk_desc_kvec(type) &&
+ ops->add_iov_frag != NULL));
+
+ if (type & PTLRPC_BULK_BUF_KIOV) {
+ OBD_ALLOC(desc,
+ offsetof(struct ptlrpc_bulk_desc,
+ bd_u.bd_kiov.bd_vec[nfrags]));
+ } else {
+ OBD_ALLOC(desc,
+ offsetof(struct ptlrpc_bulk_desc,
+ bd_u.bd_kvec.bd_kvec[nfrags]));
+ }
+
if (!desc)
return NULL;
spin_lock_init(&desc->bd_lock);
init_waitqueue_head(&desc->bd_waitq);
- desc->bd_max_iov = npages;
+ desc->bd_max_iov = nfrags;
desc->bd_iov_count = 0;
desc->bd_portal = portal;
desc->bd_type = type;
desc->bd_md_count = 0;
+ desc->bd_frag_ops = (struct ptlrpc_bulk_frag_ops *) ops;
LASSERT(max_brw > 0);
desc->bd_md_max_brw = min(max_brw, PTLRPC_BULK_OPS_COUNT);
/* PTLRPC_BULK_OPS_COUNT is the compile-time transfer limit for this
/**
* Prepare bulk descriptor for specified outgoing request \a req that
- * can fit \a npages * pages. \a type is bulk type. \a portal is where
+ * can fit \a nfrags * pages. \a type is bulk type. \a portal is where
* the bulk to be sent. Used on client-side.
* Returns pointer to newly allocatrd initialized bulk descriptor or NULL on
* error.
*/
struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
- unsigned npages, unsigned max_brw,
- unsigned type, unsigned portal)
+ unsigned nfrags, unsigned max_brw,
+ unsigned int type,
+ unsigned portal,
+ const struct ptlrpc_bulk_frag_ops
+ *ops)
{
struct obd_import *imp = req->rq_import;
struct ptlrpc_bulk_desc *desc;
ENTRY;
- LASSERT(type == BULK_PUT_SINK || type == BULK_GET_SOURCE);
- desc = ptlrpc_new_bulk(npages, max_brw, type, portal);
+ LASSERT(ptlrpc_is_bulk_op_passive(type));
+
+ desc = ptlrpc_new_bulk(nfrags, max_brw, type, portal, ops);
if (desc == NULL)
RETURN(NULL);
}
EXPORT_SYMBOL(ptlrpc_prep_bulk_imp);
-/*
- * Add a page \a page to the bulk descriptor \a desc.
- * Data to transfer in the page starts at offset \a pageoffset and
- * amount of data to transfer from the page is \a len
- */
void __ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
- struct page *page, int pageoffset, int len, int pin)
+ struct page *page, int pageoffset, int len,
+ int pin)
{
+ lnet_kiov_t *kiov;
+
LASSERT(desc->bd_iov_count < desc->bd_max_iov);
LASSERT(page != NULL);
LASSERT(pageoffset >= 0);
LASSERT(len > 0);
LASSERT(pageoffset + len <= PAGE_CACHE_SIZE);
+ LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
+
+ kiov = &BD_GET_KIOV(desc, desc->bd_iov_count);
desc->bd_nob += len;
if (pin)
page_cache_get(page);
- ptlrpc_add_bulk_page(desc, page, pageoffset, len);
+ kiov->kiov_page = page;
+ kiov->kiov_offset = pageoffset;
+ kiov->kiov_len = len;
+
+ desc->bd_iov_count++;
}
EXPORT_SYMBOL(__ptlrpc_prep_bulk_page);
-/**
- * Uninitialize and free bulk descriptor \a desc.
- * Works on bulk descriptors both from server and client side.
- */
-void __ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc, int unpin)
+int ptlrpc_prep_bulk_frag(struct ptlrpc_bulk_desc *desc,
+ void *frag, int len)
+{
+ struct kvec *iovec;
+ ENTRY;
+
+ LASSERT(desc->bd_iov_count < desc->bd_max_iov);
+ LASSERT(frag != NULL);
+ LASSERT(len > 0);
+ LASSERT(ptlrpc_is_bulk_desc_kvec(desc->bd_type));
+
+ iovec = &BD_GET_KVEC(desc, desc->bd_iov_count);
+
+ desc->bd_nob += len;
+
+ iovec->iov_base = frag;
+ iovec->iov_len = len;
+
+ desc->bd_iov_count++;
+
+ RETURN(desc->bd_nob);
+}
+EXPORT_SYMBOL(ptlrpc_prep_bulk_frag);
+
+void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *desc)
{
- int i;
ENTRY;
LASSERT(desc != NULL);
LASSERT(desc->bd_iov_count != LI_POISON); /* not freed already */
LASSERT(desc->bd_md_count == 0); /* network hands off */
LASSERT((desc->bd_export != NULL) ^ (desc->bd_import != NULL));
+ LASSERT(desc->bd_frag_ops != NULL);
- sptlrpc_enc_pool_put_pages(desc);
+ if (ptlrpc_is_bulk_desc_kiov(desc->bd_type))
+ sptlrpc_enc_pool_put_pages(desc);
if (desc->bd_export)
class_export_put(desc->bd_export);
else
class_import_put(desc->bd_import);
- if (unpin) {
- for (i = 0; i < desc->bd_iov_count ; i++)
- page_cache_release(desc->bd_iov[i].kiov_page);
- }
+ if (desc->bd_frag_ops->release_frags != NULL)
+ desc->bd_frag_ops->release_frags(desc);
+
+ if (ptlrpc_is_bulk_desc_kiov(desc->bd_type))
+ OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc,
+ bd_u.bd_kiov.bd_vec[desc->bd_max_iov]));
+ else
+ OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc,
+ bd_u.bd_kvec.bd_kvec[desc->
+ bd_max_iov]));
- OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc,
- bd_iov[desc->bd_max_iov]));
EXIT;
}
-EXPORT_SYMBOL(__ptlrpc_free_bulk);
+EXPORT_SYMBOL(ptlrpc_free_bulk);
/**
* Set server timelimit for this req, i.e. how long are we willing to wait
/**
* Allocates, initializes and adds \a num_rq requests to the pool \a pool
*/
-void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq)
+int ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq)
{
int i;
int size = 1;
spin_unlock(&pool->prp_lock);
req = ptlrpc_request_cache_alloc(GFP_NOFS);
if (!req)
- return;
+ return i;
OBD_ALLOC_LARGE(msg, size);
if (!msg) {
ptlrpc_request_cache_free(req);
- return;
+ return i;
}
req->rq_reqbuf = msg;
req->rq_reqbuf_len = size;
list_add_tail(&req->rq_list, &pool->prp_req_list);
}
spin_unlock(&pool->prp_lock);
- return;
+ return num_rq;
}
EXPORT_SYMBOL(ptlrpc_add_rqs_to_pool);
*/
struct ptlrpc_request_pool *
ptlrpc_init_rq_pool(int num_rq, int msgsize,
- void (*populate_pool)(struct ptlrpc_request_pool *, int))
+ int (*populate_pool)(struct ptlrpc_request_pool *, int))
{
struct ptlrpc_request_pool *pool;
populate_pool(pool, num_rq);
- if (list_empty(&pool->prp_req_list)) {
- /* have not allocated a single request for the pool */
- OBD_FREE(pool, sizeof(struct ptlrpc_request_pool));
- pool = NULL;
- }
return pool;
}
EXPORT_SYMBOL(ptlrpc_init_rq_pool);
spin_unlock(&pool->prp_lock);
}
-static int __ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
- __u32 version, int opcode,
- int count, __u32 *lengths, char **bufs,
- struct ptlrpc_cli_ctx *ctx)
+void ptlrpc_add_unreplied(struct ptlrpc_request *req)
{
- struct obd_import *imp = request->rq_import;
- int rc;
- ENTRY;
+ struct obd_import *imp = req->rq_import;
+ struct list_head *tmp;
+ struct ptlrpc_request *iter;
- if (unlikely(ctx))
- request->rq_cli_ctx = sptlrpc_cli_ctx_get(ctx);
- else {
- rc = sptlrpc_req_get_ctx(request);
- if (rc)
- GOTO(out_free, rc);
- }
+ assert_spin_locked(&imp->imp_lock);
+ LASSERT(list_empty(&req->rq_unreplied_list));
- sptlrpc_req_set_flavor(request, opcode);
+ /* unreplied list is sorted by xid in ascending order */
+ list_for_each_prev(tmp, &imp->imp_unreplied_list) {
+ iter = list_entry(tmp, struct ptlrpc_request,
+ rq_unreplied_list);
- rc = lustre_pack_request(request, imp->imp_msg_magic, count,
- lengths, bufs);
- if (rc) {
- LASSERT(!request->rq_pool);
- GOTO(out_ctx, rc);
- }
+ LASSERT(req->rq_xid != iter->rq_xid);
+ if (req->rq_xid < iter->rq_xid)
+ continue;
+ list_add(&req->rq_unreplied_list, &iter->rq_unreplied_list);
+ return;
+ }
+ list_add(&req->rq_unreplied_list, &imp->imp_unreplied_list);
+}
- lustre_msg_add_version(request->rq_reqmsg, version);
- request->rq_send_state = LUSTRE_IMP_FULL;
- request->rq_type = PTL_RPC_MSG_REQUEST;
+void ptlrpc_assign_next_xid_nolock(struct ptlrpc_request *req)
+{
+ req->rq_xid = ptlrpc_next_xid();
+ ptlrpc_add_unreplied(req);
+}
+
+static inline void ptlrpc_assign_next_xid(struct ptlrpc_request *req)
+{
+ spin_lock(&req->rq_import->imp_lock);
+ ptlrpc_assign_next_xid_nolock(req);
+ spin_unlock(&req->rq_import->imp_lock);
+}
- request->rq_req_cbid.cbid_fn = request_out_callback;
- request->rq_req_cbid.cbid_arg = request;
+int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
+ __u32 version, int opcode, char **bufs,
+ struct ptlrpc_cli_ctx *ctx)
+{
+ int count;
+ struct obd_import *imp;
+ __u32 *lengths;
+ int rc;
- request->rq_reply_cbid.cbid_fn = reply_in_callback;
- request->rq_reply_cbid.cbid_arg = request;
+ ENTRY;
- request->rq_reply_deadline = 0;
- request->rq_phase = RQ_PHASE_NEW;
- request->rq_next_phase = RQ_PHASE_UNDEFINED;
+ count = req_capsule_filled_sizes(&request->rq_pill, RCL_CLIENT);
+ imp = request->rq_import;
+ lengths = request->rq_pill.rc_area[RCL_CLIENT];
- request->rq_request_portal = imp->imp_client->cli_request_portal;
- request->rq_reply_portal = imp->imp_client->cli_reply_portal;
+ if (ctx != NULL) {
+ request->rq_cli_ctx = sptlrpc_cli_ctx_get(ctx);
+ } else {
+ rc = sptlrpc_req_get_ctx(request);
+ if (rc)
+ GOTO(out_free, rc);
+ }
+ sptlrpc_req_set_flavor(request, opcode);
- ptlrpc_at_set_req_timeout(request);
+ rc = lustre_pack_request(request, imp->imp_msg_magic, count,
+ lengths, bufs);
+ if (rc)
+ GOTO(out_ctx, rc);
+
+ lustre_msg_add_version(request->rq_reqmsg, version);
+ request->rq_send_state = LUSTRE_IMP_FULL;
+ request->rq_type = PTL_RPC_MSG_REQUEST;
+
+ request->rq_req_cbid.cbid_fn = request_out_callback;
+ request->rq_req_cbid.cbid_arg = request;
+
+ request->rq_reply_cbid.cbid_fn = reply_in_callback;
+ request->rq_reply_cbid.cbid_arg = request;
+
+ request->rq_reply_deadline = 0;
+ request->rq_phase = RQ_PHASE_NEW;
+ request->rq_next_phase = RQ_PHASE_UNDEFINED;
+
+ request->rq_request_portal = imp->imp_client->cli_request_portal;
+ request->rq_reply_portal = imp->imp_client->cli_reply_portal;
+
+ ptlrpc_at_set_req_timeout(request);
- request->rq_xid = ptlrpc_next_xid();
lustre_msg_set_opc(request->rq_reqmsg, opcode);
+ ptlrpc_assign_next_xid(request);
RETURN(0);
+
out_ctx:
+ LASSERT(!request->rq_pool);
sptlrpc_cli_ctx_put(request->rq_cli_ctx, 1);
out_free:
class_import_put(imp);
- return rc;
-}
-int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
- __u32 version, int opcode, char **bufs,
- struct ptlrpc_cli_ctx *ctx)
-{
- int count;
+ return rc;
- count = req_capsule_filled_sizes(&request->rq_pill, RCL_CLIENT);
- return __ptlrpc_request_bufs_pack(request, version, opcode, count,
- request->rq_pill.rc_area[RCL_CLIENT],
- bufs, ctx);
}
EXPORT_SYMBOL(ptlrpc_request_bufs_pack);
{
struct ptlrpc_request *request = NULL;
- if (pool)
- request = ptlrpc_prep_req_from_pool(pool);
+ request = ptlrpc_request_cache_alloc(GFP_NOFS);
- if (!request)
- request = ptlrpc_request_cache_alloc(GFP_NOFS);
+ if (!request && pool)
+ request = ptlrpc_prep_req_from_pool(pool);
if (request) {
ptlrpc_cli_req_init(request);
EXPORT_SYMBOL(ptlrpc_request_alloc_pack);
/**
- * Prepare request (fetched from pool \a poolif not NULL) on import \a imp
- * for operation \a opcode. Request would contain \a count buffers.
- * Sizes of buffers are described in array \a lengths and buffers themselves
- * are provided by a pointer \a bufs.
- * Returns prepared request structure pointer or NULL on error.
- */
-struct ptlrpc_request *
-ptlrpc_prep_req_pool(struct obd_import *imp,
- __u32 version, int opcode,
- int count, __u32 *lengths, char **bufs,
- struct ptlrpc_request_pool *pool)
-{
- struct ptlrpc_request *request;
- int rc;
-
- request = __ptlrpc_request_alloc(imp, pool);
- if (!request)
- return NULL;
-
- rc = __ptlrpc_request_bufs_pack(request, version, opcode, count,
- lengths, bufs, NULL);
- if (rc) {
- ptlrpc_request_free(request);
- request = NULL;
- }
- return request;
-}
-
-/**
- * Same as ptlrpc_prep_req_pool, but without pool
- */
-struct ptlrpc_request *
-ptlrpc_prep_req(struct obd_import *imp, __u32 version, int opcode, int count,
- __u32 *lengths, char **bufs)
-{
- return ptlrpc_prep_req_pool(imp, version, opcode, count, lengths, bufs,
- NULL);
-}
-
-/**
- * Allocate and initialize new request set structure.
+ * Allocate and initialize new request set structure on the current CPT.
* Returns a pointer to the newly allocated set structure or NULL on error.
*/
struct ptlrpc_request_set *ptlrpc_prep_set(void)
{
- struct ptlrpc_request_set *set;
+ struct ptlrpc_request_set *set;
+ int cpt;
ENTRY;
- OBD_ALLOC(set, sizeof *set);
+ cpt = cfs_cpt_current(cfs_cpt_table, 0);
+ OBD_CPT_ALLOC(set, cfs_cpt_table, cpt, sizeof *set);
if (!set)
RETURN(NULL);
atomic_set(&set->set_refcount, 1);
{
LASSERT(list_empty(&req->rq_set_chain));
+ if (req->rq_allow_intr)
+ set->set_allow_intr = 1;
+
/* The set takes over the caller's request reference */
list_add_tail(&req->rq_set_chain, &set->set_requests);
req->rq_set = set;
EXIT;
}
+__u64 ptlrpc_known_replied_xid(struct obd_import *imp)
+{
+ struct ptlrpc_request *req;
+
+ assert_spin_locked(&imp->imp_lock);
+ if (list_empty(&imp->imp_unreplied_list))
+ return 0;
+
+ req = list_entry(imp->imp_unreplied_list.next, struct ptlrpc_request,
+ rq_unreplied_list);
+ LASSERTF(req->rq_xid >= 1, "XID:"LPU64"\n", req->rq_xid);
+
+ if (imp->imp_known_replied_xid < req->rq_xid - 1)
+ imp->imp_known_replied_xid = req->rq_xid - 1;
+
+ return req->rq_xid - 1;
+}
+
/**
* Callback function called when client receives RPC reply for \a req.
* Returns 0 on success or error code.
struct obd_device *obd = req->rq_import->imp_obd;
int rc;
struct timeval work_start;
+ __u64 committed;
long timediff;
ENTRY;
spin_unlock(&req->rq_lock);
req->rq_nr_resend++;
- /* allocate new xid to avoid reply reconstruction */
- if (!req->rq_bulk) {
- /* new xid is already allocated for bulk in
- * ptlrpc_check_set() */
- req->rq_xid = ptlrpc_next_xid();
- DEBUG_REQ(D_RPCTRACE, req, "Allocating new xid for "
- "resend on EINPROGRESS");
- }
-
/* Readjust the timeout for current conditions */
ptlrpc_at_set_req_timeout(req);
/* delay resend to give a chance to the server to get ready.
else
req->rq_sent = now + req->rq_nr_resend;
+ /* Resend for EINPROGRESS will use a new XID */
+ spin_lock(&imp->imp_lock);
+ list_del_init(&req->rq_unreplied_list);
+ spin_unlock(&imp->imp_lock);
+
RETURN(0);
}
/*
* Replay-enabled imports return commit-status information.
*/
- if (lustre_msg_get_last_committed(req->rq_repmsg)) {
- imp->imp_peer_committed_transno =
- lustre_msg_get_last_committed(req->rq_repmsg);
- }
+ committed = lustre_msg_get_last_committed(req->rq_repmsg);
+ if (likely(committed > imp->imp_peer_committed_transno))
+ imp->imp_peer_committed_transno = committed;
ptlrpc_free_committed(imp);
static int ptlrpc_send_new_req(struct ptlrpc_request *req)
{
struct obd_import *imp = req->rq_import;
+ __u64 min_xid = 0;
int rc;
ENTRY;
LASSERT(req->rq_phase == RQ_PHASE_NEW);
+
+ /* do not try to go further if there is not enough memory in enc_pool */
+ if (req->rq_sent && req->rq_bulk != NULL)
+ if (req->rq_bulk->bd_iov_count > get_free_pages_in_pool() &&
+ pool_is_at_full_capacity())
+ RETURN(-ENOMEM);
+
if (req->rq_sent && (req->rq_sent > cfs_time_current_sec()) &&
(!req->rq_generation_set ||
req->rq_import_generation == imp->imp_generation))
spin_lock(&imp->imp_lock);
+ LASSERT(req->rq_xid != 0);
+ LASSERT(!list_empty(&req->rq_unreplied_list));
+
if (!req->rq_generation_set)
req->rq_import_generation = imp->imp_generation;
LASSERT(list_empty(&req->rq_list));
list_add_tail(&req->rq_list, &imp->imp_sending_list);
atomic_inc(&req->rq_import->imp_inflight);
+
+ /* find the known replied XID from the unreplied list, CONNECT
+ * and DISCONNECT requests are skipped to make the sanity check
+ * on server side happy. see process_req_last_xid().
+ *
+ * For CONNECT: Because replay requests have lower XID, it'll
+ * break the sanity check if CONNECT bump the exp_last_xid on
+ * server.
+ *
+ * For DISCONNECT: Since client will abort inflight RPC before
+ * sending DISCONNECT, DISCONNECT may carry an XID which higher
+ * than the inflight RPC.
+ */
+ if (!ptlrpc_req_is_connect(req) && !ptlrpc_req_is_disconnect(req))
+ min_xid = ptlrpc_known_replied_xid(imp);
spin_unlock(&imp->imp_lock);
+ lustre_msg_set_last_xid(req->rq_reqmsg, min_xid);
+
lustre_msg_set_status(req->rq_reqmsg, current_pid());
rc = sptlrpc_req_refresh_ctx(req, -1);
lustre_msg_get_opc(req->rq_reqmsg));
rc = ptl_send_rpc(req, 0);
+ if (rc == -ENOMEM) {
+ spin_lock(&imp->imp_lock);
+ if (!list_empty(&req->rq_list)) {
+ list_del_init(&req->rq_list);
+ atomic_dec(&req->rq_import->imp_inflight);
+ }
+ spin_unlock(&imp->imp_lock);
+ ptlrpc_rqphase_move(req, RQ_PHASE_NEW);
+ RETURN(rc);
+ }
if (rc) {
DEBUG_REQ(D_HA, req, "send failed (%d); expect timeout", rc);
spin_lock(&req->rq_lock);
rq_set_chain);
struct obd_import *imp = req->rq_import;
int unregistered = 0;
+ int async = 1;
int rc = 0;
+ if (req->rq_phase == RQ_PHASE_COMPLETE) {
+ list_move_tail(&req->rq_set_chain, &comp_reqs);
+ continue;
+ }
+
/* This schedule point is mainly for the ptlrpcd caller of this
* function. Most ptlrpc sets are not long-lived and unbounded
* in length, but at the least the set used by the ptlrpcd is.
*/
cond_resched();
- if (req->rq_phase == RQ_PHASE_NEW &&
- ptlrpc_send_new_req(req)) {
- force_timer_recalc = 1;
- }
+ /* If the caller requires to allow to be interpreted by force
+ * and it has really been interpreted, then move the request
+ * to RQ_PHASE_INTERPRET phase in spite of what the current
+ * phase is. */
+ if (unlikely(req->rq_allow_intr && req->rq_intr)) {
+ req->rq_status = -EINTR;
+ ptlrpc_rqphase_move(req, RQ_PHASE_INTERPRET);
+
+ /* Since it is interpreted and we have to wait for
+ * the reply to be unlinked, then use sync mode. */
+ async = 0;
+
+ GOTO(interpret, req->rq_status);
+ }
- /* delayed send - skip */
- if (req->rq_phase == RQ_PHASE_NEW && req->rq_sent)
+ if (req->rq_phase == RQ_PHASE_NEW && ptlrpc_send_new_req(req))
+ force_timer_recalc = 1;
+
+ /* delayed send - skip */
+ if (req->rq_phase == RQ_PHASE_NEW && req->rq_sent)
continue;
/* delayed resend - skip */
req->rq_sent > cfs_time_current_sec())
continue;
- if (!(req->rq_phase == RQ_PHASE_RPC ||
- req->rq_phase == RQ_PHASE_BULK ||
- req->rq_phase == RQ_PHASE_INTERPRET ||
- req->rq_phase == RQ_PHASE_UNREGISTERING ||
- req->rq_phase == RQ_PHASE_COMPLETE)) {
+ if (!(req->rq_phase == RQ_PHASE_RPC ||
+ req->rq_phase == RQ_PHASE_BULK ||
+ req->rq_phase == RQ_PHASE_INTERPRET ||
+ req->rq_phase == RQ_PHASE_UNREGISTERING)) {
DEBUG_REQ(D_ERROR, req, "bad phase %x", req->rq_phase);
LBUG();
}
ptlrpc_rqphase_move(req, req->rq_next_phase);
}
- if (req->rq_phase == RQ_PHASE_COMPLETE) {
- list_move_tail(&req->rq_set_chain, &comp_reqs);
- continue;
- }
-
if (req->rq_phase == RQ_PHASE_INTERPRET)
GOTO(interpret, req->rq_status);
spin_lock(&req->rq_lock);
req->rq_resend = 1;
spin_unlock(&req->rq_lock);
- if (req->rq_bulk) {
- __u64 old_xid;
-
- if (!ptlrpc_unregister_bulk(req, 1))
- continue;
-
- /* ensure previous bulk fails */
- old_xid = req->rq_xid;
- req->rq_xid = ptlrpc_next_xid();
- CDEBUG(D_HA, "resend bulk "
- "old x"LPU64
- " new x"LPU64"\n",
- old_xid, req->rq_xid);
- }
+
+ if (req->rq_bulk != NULL &&
+ !ptlrpc_unregister_bulk(req, 1))
+ continue;
}
/*
* rq_wait_ctx is only touched by ptlrpcd,
}
rc = ptl_send_rpc(req, 0);
+ if (rc == -ENOMEM) {
+ spin_lock(&imp->imp_lock);
+ if (!list_empty(&req->rq_list))
+ list_del_init(&req->rq_list);
+ spin_unlock(&imp->imp_lock);
+ ptlrpc_rqphase_move(req, RQ_PHASE_NEW);
+ continue;
+ }
if (rc) {
DEBUG_REQ(D_HA, req,
"send failed: rc = %d", rc);
req->rq_status = -EIO;
}
- ptlrpc_rqphase_move(req, RQ_PHASE_INTERPRET);
+ ptlrpc_rqphase_move(req, RQ_PHASE_INTERPRET);
- interpret:
- LASSERT(req->rq_phase == RQ_PHASE_INTERPRET);
+ interpret:
+ LASSERT(req->rq_phase == RQ_PHASE_INTERPRET);
- /* This moves to "unregistering" phase we need to wait for
- * reply unlink. */
- if (!unregistered && !ptlrpc_unregister_reply(req, 1)) {
- /* start async bulk unlink too */
- ptlrpc_unregister_bulk(req, 1);
- continue;
- }
+ /* This moves to "unregistering" phase we need to wait for
+ * reply unlink. */
+ if (!unregistered && !ptlrpc_unregister_reply(req, async)) {
+ /* start async bulk unlink too */
+ ptlrpc_unregister_bulk(req, 1);
+ continue;
+ }
- if (!ptlrpc_unregister_bulk(req, 1))
- continue;
+ if (!ptlrpc_unregister_bulk(req, async))
+ continue;
- /* When calling interpret receiving already should be
- * finished. */
- LASSERT(!req->rq_receiving_reply);
+ /* When calling interpret receiving already should be
+ * finished. */
+ LASSERT(!req->rq_receiving_reply);
- ptlrpc_req_interpret(env, req, req->rq_status);
+ ptlrpc_req_interpret(env, req, req->rq_status);
if (ptlrpcd_check_work(req)) {
atomic_dec(&set->set_remaining);
list_del_init(&req->rq_list);
atomic_dec(&imp->imp_inflight);
}
+ list_del_init(&req->rq_unreplied_list);
spin_unlock(&imp->imp_lock);
atomic_dec(&set->set_remaining);
"timed out for sent delay" : "timed out for slow reply"),
req->rq_sent, req->rq_real_sent);
- if (imp != NULL && obd_debug_peer_on_timeout)
- LNetCtl(IOC_LIBCFS_DEBUG_PEER, &imp->imp_connection->c_peer);
+ if (imp != NULL && obd_debug_peer_on_timeout)
+ LNetDebugPeer(imp->imp_connection->c_peer);
ptlrpc_unregister_reply(req, async_unlink);
ptlrpc_unregister_bulk(req, async_unlink);
struct ptlrpc_request *req =
list_entry(tmp, struct ptlrpc_request, rq_set_chain);
+ if (req->rq_intr)
+ continue;
+
if (req->rq_phase != RQ_PHASE_RPC &&
- req->rq_phase != RQ_PHASE_UNREGISTERING)
+ req->rq_phase != RQ_PHASE_UNREGISTERING &&
+ !req->rq_allow_intr)
continue;
ptlrpc_mark_interrupted(req);
CDEBUG(D_RPCTRACE, "set %p going to sleep for %d seconds\n",
set, timeout);
- if (timeout == 0 && !signal_pending(current))
- /*
- * No requests are in-flight (ether timed out
- * or delayed), so we can allow interrupts.
- * We still want to block for a limited time,
- * so we allow interrupts during the timeout.
- */
- lwi = LWI_TIMEOUT_INTR_ALL(cfs_time_seconds(1),
- ptlrpc_expired_set,
- ptlrpc_interrupted_set, set);
+ if ((timeout == 0 && !signal_pending(current)) ||
+ set->set_allow_intr)
+ /* No requests are in-flight (ether timed out
+ * or delayed), so we can allow interrupts.
+ * We still want to block for a limited time,
+ * so we allow interrupts during the timeout. */
+ lwi = LWI_TIMEOUT_INTR_ALL(
+ cfs_time_seconds(timeout ? timeout : 1),
+ ptlrpc_expired_set,
+ ptlrpc_interrupted_set, set);
else
/*
* At least one request is in flight, so no
/* LU-769 - if we ignored the signal because it was already
* pending when we started, we need to handle it now or we risk
* it being ignored forever */
- if (rc == -ETIMEDOUT && !lwi.lwi_allow_intr &&
+ if (rc == -ETIMEDOUT &&
+ (!lwi.lwi_allow_intr || set->set_allow_intr) &&
signal_pending(current)) {
sigset_t blocked_sigs =
cfs_block_sigsinv(LUSTRE_FATAL_SIGS);
if (!locked)
spin_lock(&request->rq_import->imp_lock);
list_del_init(&request->rq_replay_list);
+ list_del_init(&request->rq_unreplied_list);
if (!locked)
spin_unlock(&request->rq_import->imp_lock);
}
request->rq_import = NULL;
}
if (request->rq_bulk != NULL)
- ptlrpc_free_bulk_pin(request->rq_bulk);
+ ptlrpc_free_bulk(request->rq_bulk);
if (request->rq_reqbuf != NULL || request->rq_clrbuf != NULL)
sptlrpc_cli_free_reqbuf(request);
req->rq_resend = 1;
req->rq_net_err = 0;
req->rq_timedout = 0;
- if (req->rq_bulk) {
- __u64 old_xid = req->rq_xid;
- /* ensure previous bulk fails */
- req->rq_xid = ptlrpc_next_xid();
- CDEBUG(D_HA, "resend bulk old x"LPU64" new x"LPU64"\n",
- old_xid, req->rq_xid);
- }
ptlrpc_client_wake_req(req);
spin_unlock(&req->rq_lock);
}
lustre_msg_add_flags(req->rq_reqmsg, MSG_REPLAY);
+ spin_lock(&req->rq_lock);
+ req->rq_resend = 0;
+ spin_unlock(&req->rq_lock);
+
LASSERT(imp->imp_replayable);
/* Balanced in ptlrpc_free_committed, usually. */
ptlrpc_request_addref(req);
ENTRY;
atomic_dec(&imp->imp_replay_inflight);
- if (!ptlrpc_client_replied(req)) {
- CERROR("request replay timed out, restarting recovery\n");
- GOTO(out, rc = -ETIMEDOUT);
- }
+ /* Note: if it is bulk replay (MDS-MDS replay), then even if
+ * server got the request, but bulk transfer timeout, let's
+ * replay the bulk req again */
+ if (!ptlrpc_client_replied(req) ||
+ (req->rq_bulk != NULL &&
+ lustre_msg_get_status(req->rq_repmsg) == -ETIMEDOUT)) {
+ DEBUG_REQ(D_ERROR, req, "request replay timed out.\n");
+ GOTO(out, rc = -ETIMEDOUT);
+ }
if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR &&
(lustre_msg_get_status(req->rq_repmsg) == -ENOTCONN ||
DEBUG_REQ(D_ERROR, req, "status %d, old was %d",
lustre_msg_get_status(req->rq_repmsg),
aa->praa_old_status);
+
+ /* Note: If the replay fails for MDT-MDT recovery, let's
+ * abort all of the following requests in the replay
+ * and sending list, because MDT-MDT update requests
+ * are dependent on each other, see LU-7039 */
+ if (imp->imp_connect_flags_orig & OBD_CONNECT_MDS_MDS) {
+ struct ptlrpc_request *free_req;
+ struct ptlrpc_request *tmp;
+
+ spin_lock(&imp->imp_lock);
+ list_for_each_entry_safe(free_req, tmp,
+ &imp->imp_replay_list,
+ rq_replay_list) {
+ ptlrpc_free_request(free_req);
+ }
+
+ list_for_each_entry_safe(free_req, tmp,
+ &imp->imp_committed_list,
+ rq_replay_list) {
+ ptlrpc_free_request(free_req);
+ }
+
+ list_for_each_entry_safe(free_req, tmp,
+ &imp->imp_delayed_list,
+ rq_list) {
+ spin_lock(&free_req->rq_lock);
+ free_req->rq_err = 1;
+ free_req->rq_status = -EIO;
+ ptlrpc_client_wake_req(free_req);
+ spin_unlock(&free_req->rq_lock);
+ }
+
+ list_for_each_entry_safe(free_req, tmp,
+ &imp->imp_sending_list,
+ rq_list) {
+ spin_lock(&free_req->rq_lock);
+ free_req->rq_err = 1;
+ free_req->rq_status = -EIO;
+ ptlrpc_client_wake_req(free_req);
+ spin_unlock(&free_req->rq_lock);
+ }
+ spin_unlock(&imp->imp_lock);
+ }
} else {
/* Put it back for re-replay. */
lustre_msg_set_status(req->rq_repmsg, aa->praa_old_status);
atomic_inc(&req->rq_import->imp_replay_inflight);
ptlrpc_request_addref(req); /* ptlrpcd needs a ref */
- ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
+ ptlrpcd_add_req(req);
RETURN(0);
}
}
/**
+ * If request has a new allocated XID (new request or EINPROGRESS resend),
+ * use this XID as matchbits of bulk, otherwise allocate a new matchbits for
+ * request to ensure previous bulk fails and avoid problems with lost replies
+ * and therefore several transfers landing into the same buffer from different
+ * sending attempts.
+ */
+void ptlrpc_set_bulk_mbits(struct ptlrpc_request *req)
+{
+ struct ptlrpc_bulk_desc *bd = req->rq_bulk;
+
+ LASSERT(bd != NULL);
+
+ if (!req->rq_resend) {
+ /* this request has a new xid, just use it as bulk matchbits */
+ req->rq_mbits = req->rq_xid;
+
+ } else { /* needs to generate a new matchbits for resend */
+ __u64 old_mbits = req->rq_mbits;
+
+ if ((bd->bd_import->imp_connect_data.ocd_connect_flags &
+ OBD_CONNECT_BULK_MBITS) != 0)
+ req->rq_mbits = ptlrpc_next_xid();
+ else /* old version transfers rq_xid to peer as matchbits */
+ req->rq_mbits = req->rq_xid = ptlrpc_next_xid();
+
+ CDEBUG(D_HA, "resend bulk old x"LPU64" new x"LPU64"\n",
+ old_mbits, req->rq_mbits);
+ }
+
+ /* For multi-bulk RPCs, rq_mbits is the last mbits needed for bulks so
+ * that server can infer the number of bulks that were prepared,
+ * see LU-1431 */
+ req->rq_mbits += ((bd->bd_iov_count + LNET_MAX_IOV - 1) /
+ LNET_MAX_IOV) - 1;
+}
+
+/**
* Get a glimpse at what next xid value might have been.
* Returns possible next xid.
*/
req->rq_xid = ptlrpc_next_xid();
req->rq_import_generation = req->rq_import->imp_generation;
- ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
+ ptlrpcd_add_req(req);
}
static int work_interpreter(const struct lu_env *env,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff