/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (C) 2001-2003 Cluster File Systems, Inc.
- * Author: Peter J. Braam <braam@clusterfs.com>
- * Author: Phil Schwan <phil@clusterfs.com>
+ * GPL HEADER START
*
- * This file is part of the Lustre file system, http://www.lustre.org
- * Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
- * You may have signed or agreed to another license before downloading
- * this software. If so, you are bound by the terms and conditions
- * of that agreement, and the following does not apply to you. See the
- * LICENSE file included with this distribution for more information.
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
*
- * If you did not agree to a different license, then this copy of Lustre
- * is open source software; you can redistribute it and/or modify it
- * under the terms of version 2 of the GNU General Public License as
- * published by the Free Software Foundation.
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
*
- * In either case, Lustre is distributed in the hope that it will be
- * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- * license text for more details.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
*
- * Storage Target Handling functions
- * Lustre Object Server Module (OST)
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
*
- * This server is single threaded at present (but can easily be multi
- * threaded). For testing and management it is treated as an
- * obd_device, although it does not export a full OBD method table
- * (the requests are coming in over the wire, so object target
- * modules do not have a full method table.)
+ * GPL HEADER END
+ */
+/*
+ * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/ost/ost_handler.c
+ *
+ * Author: Peter J. Braam <braam@clusterfs.com>
+ * Author: Phil Schwan <phil@clusterfs.com>
*/
#ifndef EXPORT_SYMTAB
#define DEBUG_SUBSYSTEM S_OST
#include <linux/module.h>
+#include <obd_cksum.h>
#include <obd_ost.h>
#include <lustre_net.h>
#include <lustre_dlm.h>
#include <lustre_debug.h>
#include <linux/init.h>
#include <lprocfs_status.h>
-#include <lustre_commit_confd.h>
#include <libcfs/list.h>
#include <lustre_quota.h>
#include "ost_internal.h"
if (!ack_lock->mode)
break;
/* XXX not even calling target_send_reply in some cases... */
- ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode);
+ ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
}
}
struct obd_trans_info *oti)
{
struct ost_body *body, *repbody;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ int rc;
ENTRY;
body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
RETURN (-EFAULT);
ldlm_request_cancel(req, dlm, 0);
}
-
+
rc = lustre_pack_reply(req, 2, size, NULL);
if (rc)
RETURN(rc);
if (body->oa.o_valid & OBD_MD_FLCOOKIE)
- oti->oti_logcookies = obdo_logcookie(&body->oa);
+ oti->oti_logcookies = &body->oa.o_lcookie;
repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
sizeof(*repbody));
memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
{
struct ost_body *body, *repbody;
struct obd_info oinfo = { { { 0 } } };
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ int rc;
ENTRY;
body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
static int ost_statfs(struct ptlrpc_request *req)
{
struct obd_statfs *osfs;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
+ int rc;
ENTRY;
rc = lustre_pack_reply(req, 2, size, NULL);
struct obd_trans_info *oti)
{
struct ost_body *body, *repbody;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ int rc;
ENTRY;
body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
sizeof(*repbody));
memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
- oti->oti_logcookies = obdo_logcookie(&repbody->oa);
+ oti->oti_logcookies = &repbody->oa.o_lcookie;
req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
//obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
RETURN(0);
struct lustre_handle *lh)
{
int flags;
- struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0} };
+ struct ldlm_res_id res_id;
ldlm_policy_data_t policy;
__u64 start;
__u64 finis;
ENTRY;
+ osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
LASSERT(!lustre_handle_is_used(lh));
if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
LDLM_EXTENT, &policy, LCK_PW, &flags,
ldlm_blocking_ast, ldlm_completion_ast,
- ldlm_glimpse_ast, NULL, 0, NULL, lh));
+ ldlm_glimpse_ast, NULL, 0, NULL,
+ NULL, lh));
}
/*
{
struct obd_info oinfo = { { { 0 } } };
struct ost_body *body, *repbody;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ int rc;
struct lustre_handle lh = {0,};
ENTRY;
if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
- RETURN(-EINVAL);
+ RETURN(-EPROTO);
rc = lustre_pack_reply(req, 2, size, NULL);
if (rc)
{
struct ost_body *body, *repbody;
struct lustre_capa *capa = NULL;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ int rc;
ENTRY;
body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
struct obd_trans_info *oti)
{
struct ost_body *body, *repbody;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
+ int rc;
struct obd_info oinfo = { { { 0 } } };
ENTRY;
RETURN(1);
}
-static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo,
- struct niobuf_remote *rnb, int nrnb,
- struct niobuf_remote **pp_rnbp)
-{
- /* Copy a remote niobuf, splitting it into page-sized chunks
- * and setting ioo[i].ioo_bufcnt accordingly */
- struct niobuf_remote *pp_rnb;
- int i;
- int j;
- int page;
- int rnbidx = 0;
- int npages = 0;
-
- /*
- * array of sufficient size already preallocated by caller
- */
- LASSERT(pp_rnbp != NULL);
- LASSERT(*pp_rnbp != NULL);
-
- /* first count and check the number of pages required */
- for (i = 0; i < nioo; i++)
- for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) {
- obd_off offset = rnb[rnbidx].offset;
- obd_off p0 = offset >> CFS_PAGE_SHIFT;
- obd_off pn = (offset + rnb[rnbidx].len - 1) >>
- CFS_PAGE_SHIFT;
-
- LASSERT(rnbidx < nrnb);
-
- npages += (pn + 1 - p0);
-
- if (rnb[rnbidx].len == 0) {
- CERROR("zero len BRW: obj %d objid "LPX64
- " buf %u\n", i, ioo[i].ioo_id, j);
- return -EINVAL;
- }
- if (j > 0 &&
- rnb[rnbidx].offset <= rnb[rnbidx-1].offset) {
- CERROR("unordered BRW: obj %d objid "LPX64
- " buf %u offset "LPX64" <= "LPX64"\n",
- i, ioo[i].ioo_id, j, rnb[rnbidx].offset,
- rnb[rnbidx].offset);
- return -EINVAL;
- }
- }
-
- LASSERT(rnbidx == nrnb);
-
- if (npages == nrnb) { /* all niobufs are for single pages */
- *pp_rnbp = rnb;
- return npages;
- }
-
- pp_rnb = *pp_rnbp;
-
- /* now do the actual split */
- page = rnbidx = 0;
- for (i = 0; i < nioo; i++) {
- int obj_pages = 0;
-
- for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
- obd_off off = rnb[rnbidx].offset;
- int nob = rnb[rnbidx].len;
-
- LASSERT(rnbidx < nrnb);
- do {
- obd_off poff = off & ~CFS_PAGE_MASK;
- int pnob = (poff + nob > CFS_PAGE_SIZE) ?
- PAGE_SIZE - poff : nob;
-
- LASSERT(page < npages);
- pp_rnb[page].len = pnob;
- pp_rnb[page].offset = off;
- pp_rnb[page].flags = rnb[rnbidx].flags;
-
- CDEBUG(0, " obj %d id "LPX64
- "page %d(%d) "LPX64" for %d, flg %x\n",
- i, ioo[i].ioo_id, obj_pages, page,
- pp_rnb[page].offset, pp_rnb[page].len,
- pp_rnb[page].flags);
- page++;
- obj_pages++;
-
- off += pnob;
- nob -= pnob;
- } while (nob > 0);
- LASSERT(nob == 0);
- }
- ioo[i].ioo_bufcnt = obj_pages;
- }
- LASSERT(page == npages);
-
- return npages;
-}
-
static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
cksum_type_t cksum_type)
{
/* corrupt the data after we compute the checksum, to
* simulate an OST->client data error */
if (i == 0 && opc == OST_READ &&
- OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND))
+ OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
memcpy(ptr, "bad4", min(4, len));
+ /* nobody should use corrupted page again */
+ ClearPageUptodate(page);
+ }
kunmap(page);
}
return cksum;
}
-/*
- * populate @nio by @nrpages pages from per-thread page pool
- */
-static void ost_nio_pages_get(struct ptlrpc_request *req,
- struct niobuf_local *nio, int nrpages)
-{
- int i;
- struct ost_thread_local_cache *tls;
-
- ENTRY;
-
- LASSERT(nrpages <= OST_THREAD_POOL_SIZE);
- LASSERT(req != NULL);
- LASSERT(req->rq_svc_thread != NULL);
-
- tls = ost_tls(req);
- LASSERT(tls != NULL);
-
- memset(nio, 0, nrpages * sizeof *nio);
- for (i = 0; i < nrpages; ++ i) {
- struct page *page;
-
- page = tls->page[i];
- LASSERT(page != NULL);
- POISON_PAGE(page, 0xf1);
- nio[i].page = page;
- LL_CDEBUG_PAGE(D_INFO, page, "%d\n", i);
- }
- EXIT;
-}
-
-/*
- * Dual for ost_nio_pages_get(). Poison pages in pool for debugging
- */
-static void ost_nio_pages_put(struct ptlrpc_request *req,
- struct niobuf_local *nio, int nrpages)
-{
- int i;
-
- ENTRY;
-
- LASSERT(nrpages <= OST_THREAD_POOL_SIZE);
-
- for (i = 0; i < nrpages; ++ i)
- POISON_PAGE(nio[i].page, 0xf2);
- EXIT;
-}
-
static int ost_brw_lock_get(int mode, struct obd_export *exp,
struct obd_ioobj *obj, struct niobuf_remote *nb,
struct lustre_handle *lh)
{
int flags = 0;
int nrbufs = obj->ioo_bufcnt;
- struct ldlm_res_id res_id = { .name = { obj->ioo_id, 0,
- obj->ioo_gr, 0} };
+ struct ldlm_res_id res_id;
ldlm_policy_data_t policy;
int i;
-
ENTRY;
+ osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
LASSERT(mode == LCK_PR || mode == LCK_PW);
LASSERT(!lustre_handle_is_used(lh));
RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
LDLM_EXTENT, &policy, mode, &flags,
ldlm_blocking_ast, ldlm_completion_ast,
- ldlm_glimpse_ast, NULL, 0, NULL, lh));
+ ldlm_glimpse_ast, NULL, 0, NULL,
+ NULL, lh));
}
static void ost_brw_lock_put(int mode,
}
static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj,
- struct niobuf_remote *nb, ldlm_mode_t mode)
+ struct niobuf_remote *nb, struct obdo *oa,
+ ldlm_mode_t mode)
{
- struct ldlm_res_id res_id = { .name = { obj->ioo_id, 0,
- obj->ioo_gr, 0} };
+ struct ldlm_res_id res_id;
int nrbufs = obj->ioo_bufcnt;
struct ost_prolong_data opd;
-
ENTRY;
+ osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
+
opd.opd_mode = mode;
opd.opd_exp = exp;
opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
opd.opd_policy.l_extent.end);
+
+ if (oa->o_valid & OBD_MD_FLHANDLE) {
+ struct ldlm_lock *lock;
+
+ lock = ldlm_handle2lock(&oa->o_handle);
+ if (lock != NULL) {
+ ost_prolong_locks_iter(lock, &opd);
+ LDLM_LOCK_PUT(lock);
+ EXIT;
+ return;
+ }
+ }
+
ldlm_resource_iterate(exp->exp_obd->obd_namespace, &res_id,
ost_prolong_locks_iter, &opd);
+
+ EXIT;
}
static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
{
- struct ptlrpc_bulk_desc *desc;
+ struct ptlrpc_bulk_desc *desc = NULL;
struct obd_export *exp = req->rq_export;
struct niobuf_remote *remote_nb;
- struct niobuf_remote *pp_rnb = NULL;
struct niobuf_local *local_nb;
struct obd_ioobj *ioo;
struct ost_body *body, *repbody;
struct lustre_capa *capa = NULL;
struct l_wait_info lwi;
struct lustre_handle lockh = { 0 };
- int size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
- int niocount, npages, nob = 0, rc, i;
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ int objcount, niocount, npages, nob = 0, rc, i;
int no_reply = 0;
ENTRY;
GOTO(out, rc = -EFAULT);
}
+ objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
+ sizeof(*ioo);
+ if (objcount == 0) {
+ CERROR("Missing/short ioobj\n");
+ GOTO(out, rc = -EFAULT);
+ }
+ if (objcount > 1) {
+ CERROR("too many ioobjs (%d)\n", objcount);
+ GOTO(out, rc = -EFAULT);
+ }
+
ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*ioo),
lustre_swab_obd_ioobj);
if (ioo == NULL) {
* ost_thread_init().
*/
local_nb = ost_tls(req)->local;
- pp_rnb = ost_tls(req)->remote;
-
- /* FIXME all niobuf splitting should be done in obdfilter if needed */
- /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
- npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
- if (npages < 0)
- GOTO(out, rc = npages);
-
- LASSERT(npages <= OST_THREAD_POOL_SIZE);
-
- ost_nio_pages_get(req, local_nb, npages);
- desc = ptlrpc_prep_bulk_exp(req, npages,
- BULK_PUT_SOURCE, OST_BULK_PORTAL);
- if (desc == NULL)
- GOTO(out, rc = -ENOMEM);
-
- rc = ost_brw_lock_get(LCK_PR, exp, ioo, pp_rnb, &lockh);
+ rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
if (rc != 0)
GOTO(out_bulk, rc);
- /*
+ /*
* If getting the lock took more time than
* client was willing to wait, drop it. b=11330
*/
- if (cfs_time_current_sec() > req->rq_arrival_time.tv_sec + obd_timeout ||
+ if (cfs_time_current_sec() > req->rq_deadline ||
OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
no_reply = 1;
CERROR("Dropping timed-out read from %s because locking"
- "object "LPX64" took %ld seconds.\n",
+ "object "LPX64" took %ld seconds (limit was %ld).\n",
libcfs_id2str(req->rq_peer), ioo->ioo_id,
- cfs_time_current_sec() - req->rq_arrival_time.tv_sec);
+ cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
+ req->rq_deadline - req->rq_arrival_time.tv_sec);
GOTO(out_lock, rc = -ETIMEDOUT);
}
- rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1,
- ioo, npages, pp_rnb, local_nb, oti, capa);
+ npages = OST_THREAD_POOL_SIZE;
+ rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
+ remote_nb, &npages, local_nb, oti, capa);
if (rc != 0)
GOTO(out_lock, rc);
- ost_prolong_locks(exp, ioo, pp_rnb, LCK_PW | LCK_PR);
+ desc = ptlrpc_prep_bulk_exp(req, npages,
+ BULK_PUT_SOURCE, OST_BULK_PORTAL);
+ if (desc == NULL) /* XXX: check all cleanup stuff */
+ GOTO(out, rc = -ENOMEM);
+
+ ost_prolong_locks(exp, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR);
nob = 0;
for (i = 0; i < npages; i++) {
break;
}
- LASSERTF(page_rc <= pp_rnb[i].len, "page_rc (%d) > "
- "pp_rnb[%d].len (%d)\n", page_rc, i, pp_rnb[i].len);
nob += page_rc;
if (page_rc != 0) { /* some data! */
LASSERT (local_nb[i].page != NULL);
ptlrpc_prep_bulk_page(desc, local_nb[i].page,
- pp_rnb[i].offset & ~CFS_PAGE_MASK,
+ local_nb[i].offset & ~CFS_PAGE_MASK,
page_rc);
}
- if (page_rc != pp_rnb[i].len) { /* short read */
+ if (page_rc != local_nb[i].len) { /* short read */
/* All subsequent pages should be 0 */
while(++i < npages)
LASSERT(local_nb[i].rc == 0);
}
if (rc == 0) {
- lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
- ost_bulk_timeout, desc);
- rc = l_wait_event(desc->bd_waitq,
- !ptlrpc_bulk_active(desc) ||
- exp->exp_failed, &lwi);
- LASSERT(rc == 0 || rc == -ETIMEDOUT);
+ time_t start = cfs_time_current_sec();
+ do {
+ long timeoutl = req->rq_deadline -
+ cfs_time_current_sec();
+ cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
+ CFS_TICK : cfs_time_seconds(timeoutl);
+ lwi = LWI_TIMEOUT_INTERVAL(timeout,
+ cfs_time_seconds(1),
+ ost_bulk_timeout,
+ desc);
+ rc = l_wait_event(desc->bd_waitq,
+ !ptlrpc_bulk_active(desc) ||
+ exp->exp_failed, &lwi);
+ LASSERT(rc == 0 || rc == -ETIMEDOUT);
+ /* Wait again if we changed deadline */
+ } while ((rc == -ETIMEDOUT) &&
+ (req->rq_deadline > cfs_time_current_sec()));
+
if (rc == -ETIMEDOUT) {
- DEBUG_REQ(D_ERROR, req, "timeout on bulk PUT");
+ DEBUG_REQ(D_ERROR, req,
+ "timeout on bulk PUT after %ld%+lds",
+ req->rq_deadline - start,
+ cfs_time_current_sec() -
+ req->rq_deadline);
ptlrpc_abort_bulk(desc);
} else if (exp->exp_failed) {
DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
}
/* Must commit after prep above in all cases */
- rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1,
- ioo, npages, local_nb, oti, rc);
-
- ost_nio_pages_put(req, local_nb, npages);
+ rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
+ remote_nb, npages, local_nb, oti, rc);
if (rc == 0) {
repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
}
out_lock:
- ost_brw_lock_put(LCK_PR, ioo, pp_rnb, &lockh);
+ ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
out_bulk:
- ptlrpc_free_bulk(desc);
+ if (desc)
+ ptlrpc_free_bulk(desc);
out:
LASSERT(rc <= 0);
if (rc == 0) {
req->rq_status = nob;
+ ptlrpc_lprocfs_brw(req, nob);
target_committed_to_req(req);
ptlrpc_reply(req);
} else if (!no_reply) {
req->rq_status = rc;
ptlrpc_error(req);
} else {
- if (req->rq_reply_state != NULL) {
- /* reply out callback would free */
- ptlrpc_rs_decref(req->rq_reply_state);
- req->rq_reply_state = NULL;
- }
+ /* reply out callback would free */
+ ptlrpc_req_drop_rs(req);
CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
"client will retry\n",
exp->exp_obd->obd_name,
static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
{
- struct ptlrpc_bulk_desc *desc;
+ struct ptlrpc_bulk_desc *desc = NULL;
struct obd_export *exp = req->rq_export;
struct niobuf_remote *remote_nb;
- struct niobuf_remote *pp_rnb;
struct niobuf_local *local_nb;
struct obd_ioobj *ioo;
struct ost_body *body, *repbody;
struct lustre_handle lockh = {0};
struct lustre_capa *capa = NULL;
__u32 *rcs;
- int size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
int objcount, niocount, npages;
int rc, swab, i, j;
obd_count client_cksum = 0, server_cksum = 0;
cksum_type_t cksum_type = OBD_CKSUM_CRC32;
- int no_reply = 0;
+ int no_reply = 0;
ENTRY;
req->rq_bulk_write = 1;
rc = lustre_pack_reply(req, 3, size, NULL);
if (rc != 0)
GOTO(out, rc);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
niocount * sizeof(*rcs));
* ost_thread_init().
*/
local_nb = ost_tls(req)->local;
- pp_rnb = ost_tls(req)->remote;
- /* FIXME all niobuf splitting should be done in obdfilter if needed */
- /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
- npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
- if (npages < 0)
- GOTO(out, rc = npages);
-
- LASSERT(npages <= OST_THREAD_POOL_SIZE);
-
- ost_nio_pages_get(req, local_nb, npages);
-
- desc = ptlrpc_prep_bulk_exp(req, npages,
- BULK_GET_SINK, OST_BULK_PORTAL);
- if (desc == NULL)
- GOTO(out, rc = -ENOMEM);
-
- rc = ost_brw_lock_get(LCK_PW, exp, ioo, pp_rnb, &lockh);
+ rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
if (rc != 0)
GOTO(out_bulk, rc);
- /*
+ /*
* If getting the lock took more time than
* client was willing to wait, drop it. b=11330
*/
- if (cfs_time_current_sec() > req->rq_arrival_time.tv_sec + obd_timeout ||
+ if (cfs_time_current_sec() > req->rq_deadline ||
OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
no_reply = 1;
- CERROR("Dropping timed-out write from %s because locking"
- "object "LPX64" took %ld seconds.\n",
+ CERROR("Dropping timed-out write from %s because locking "
+ "object "LPX64" took %ld seconds (limit was %ld).\n",
libcfs_id2str(req->rq_peer), ioo->ioo_id,
- cfs_time_current_sec() - req->rq_arrival_time.tv_sec);
+ cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
+ req->rq_deadline - req->rq_arrival_time.tv_sec);
GOTO(out_lock, rc = -ETIMEDOUT);
}
- ost_prolong_locks(exp, ioo, pp_rnb, LCK_PW);
+ ost_prolong_locks(exp, ioo, remote_nb,&body->oa, LCK_PW);
/* obd_preprw clobbers oa->valid, so save what we need */
if (body->oa.o_valid & OBD_MD_FLCKSUM) {
if (body->oa.o_valid & OBD_MD_FLFLAGS)
cksum_type = cksum_type_unpack(body->oa.o_flags);
}
-
+
/* Because we already sync grant info with client when reconnect,
- * grant info will be cleared for resent req, then fed_grant and
- * total_grant will not be modified in following preprw_write */
+ * grant info will be cleared for resent req, then fed_grant and
+ * total_grant will not be modified in following preprw_write */
if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
body->oa.o_valid &= ~OBD_MD_FLGRANT;
}
+ npages = OST_THREAD_POOL_SIZE;
rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
- ioo, npages, pp_rnb, local_nb, oti, capa);
+ ioo, remote_nb, &npages, local_nb, oti, capa);
if (rc != 0)
GOTO(out_lock, rc);
+ desc = ptlrpc_prep_bulk_exp(req, npages,
+ BULK_GET_SINK, OST_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(out, rc = -ENOMEM);
+
/* NB Having prepped, we must commit... */
for (i = 0; i < npages; i++)
ptlrpc_prep_bulk_page(desc, local_nb[i].page,
- pp_rnb[i].offset & ~CFS_PAGE_MASK,
- pp_rnb[i].len);
+ local_nb[i].offset & ~CFS_PAGE_MASK,
+ local_nb[i].len);
/* Check if client was evicted while we were doing i/o before touching
network */
else
rc = ptlrpc_start_bulk_transfer (desc);
if (rc == 0) {
- lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 2, HZ,
- ost_bulk_timeout, desc);
- rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
- desc->bd_export->exp_failed, &lwi);
- LASSERT(rc == 0 || rc == -ETIMEDOUT);
+ time_t start = cfs_time_current_sec();
+ do {
+ long timeoutl = req->rq_deadline -
+ cfs_time_current_sec();
+ cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
+ CFS_TICK : cfs_time_seconds(timeoutl);
+ lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
+ ost_bulk_timeout, desc);
+ rc = l_wait_event(desc->bd_waitq,
+ !ptlrpc_bulk_active(desc) ||
+ desc->bd_export->exp_failed, &lwi);
+ LASSERT(rc == 0 || rc == -ETIMEDOUT);
+ /* Wait again if we changed deadline */
+ } while ((rc == -ETIMEDOUT) &&
+ (req->rq_deadline > cfs_time_current_sec()));
+
if (rc == -ETIMEDOUT) {
- DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
+ DEBUG_REQ(D_ERROR, req,
+ "timeout on bulk GET after %ld%+lds",
+ req->rq_deadline - start,
+ cfs_time_current_sec() -
+ req->rq_deadline);
ptlrpc_abort_bulk(desc);
} else if (desc->bd_export->exp_failed) {
DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
}
/* Must commit after prep above in all cases */
- rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa,
- objcount, ioo, npages, local_nb, oti, rc);
+ rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
+ remote_nb, npages, local_nb, oti, rc);
if (unlikely(client_cksum != server_cksum && rc == 0)) {
int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
via = " via ";
router = libcfs_nid2str(desc->bd_sender);
}
-
+
LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
"%s%s%s inum "LPU64"/"LPU64" object "
LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
body->oa.o_id,
body->oa.o_valid & OBD_MD_FLGROUP ?
body->oa.o_gr : (__u64)0,
- pp_rnb[0].offset,
- pp_rnb[npages-1].offset+pp_rnb[npages-1].len
- - 1 );
+ local_nb[0].offset,
+ local_nb[npages-1].offset +
+ local_nb[npages-1].len - 1 );
CERROR("client csum %x, original server csum %x, "
"server csum now %x\n",
client_cksum, server_cksum, new_cksum);
}
- ost_nio_pages_put(req, local_nb, npages);
-
if (rc == 0) {
+ int nob = 0;
+
/* set per-requested niobuf return codes */
for (i = j = 0; i < niocount; i++) {
- int nob = remote_nb[i].len;
+ int len = remote_nb[i].len;
+ nob += len;
rcs[i] = 0;
do {
LASSERT(j < npages);
if (local_nb[j].rc < 0)
rcs[i] = local_nb[j].rc;
- nob -= pp_rnb[j].len;
+ len -= local_nb[j].len;
j++;
- } while (nob > 0);
- LASSERT(nob == 0);
+ } while (len > 0);
+ LASSERT(len == 0);
}
LASSERT(j == npages);
+ ptlrpc_lprocfs_brw(req, nob);
}
out_lock:
- ost_brw_lock_put(LCK_PW, ioo, pp_rnb, &lockh);
+ ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
out_bulk:
- ptlrpc_free_bulk(desc);
+ if (desc)
+ ptlrpc_free_bulk(desc);
out:
if (rc == 0) {
oti_to_request(oti, req);
req->rq_status = rc;
ptlrpc_error(req);
} else {
- if (req->rq_reply_state != NULL) {
- /* reply out callback would free */
- ptlrpc_rs_decref(req->rq_reply_state);
- req->rq_reply_state = NULL;
- }
+ /* reply out callback would free */
+ ptlrpc_req_drop_rs(req);
CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
"client will retry\n",
exp->exp_obd->obd_name,
if (vallen)
val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, 0);
- if (KEY_IS("evict_by_nid")) {
+ if (KEY_IS(KEY_EVICT_BY_NID)) {
if (val && vallen)
obd_export_evict_by_nid(exp->exp_obd, val);
}
keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
- rc = obd_get_info(exp, keylen, key, &replylen, NULL);
+ rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
if (rc)
RETURN(rc);
RETURN(-ENOMEM);
/* call again to fill in the reply buffer */
- rc = obd_get_info(exp, keylen, key, &replylen, reply);
+ rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
lustre_msg_set_status(req->rq_repmsg, 0);
RETURN(rc);
static int ost_handle_quotactl(struct ptlrpc_request *req)
{
struct obd_quotactl *oqctl, *repoqc;
- int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) };
+ __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) };
+ int rc;
ENTRY;
oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl),
}
oti_init(oti, req);
-
+
rc = ost_msg_check_version(req->rq_reqmsg);
if (rc)
RETURN(rc);
if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
RETURN(0);
rc = llog_origin_handle_cancel(req);
+ if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
+ RETURN(0);
req->rq_status = rc;
rc = req_capsule_server_pack(&req->rq_pill);
if (rc)
*/
static void ost_thread_done(struct ptlrpc_thread *thread)
{
- int i;
struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
* Storage */
*/
tls = thread->t_data;
if (tls != NULL) {
- for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) {
- if (tls->page[i] != NULL)
- OBD_PAGE_FREE(tls->page[i]);
- }
OBD_FREE_PTR(tls);
thread->t_data = NULL;
}
*/
static int ost_thread_init(struct ptlrpc_thread *thread)
{
- int result;
- int i;
struct ost_thread_local_cache *tls;
ENTRY;
LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
OBD_ALLOC_PTR(tls);
- if (tls != NULL) {
- result = 0;
- thread->t_data = tls;
- /*
- * populate pool
- */
- for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) {
- OBD_PAGE_ALLOC(tls->page[i], OST_THREAD_POOL_GFP);
- if (tls->page[i] == NULL) {
- ost_thread_done(thread);
- result = -ENOMEM;
- break;
- }
- }
- } else
- result = -ENOMEM;
- RETURN(result);
+ if (tls == NULL)
+ RETURN(-ENOMEM);
+ thread->t_data = tls;
+ RETURN(0);
}
#define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
if (oss_num_threads) {
/* If oss_num_threads is set, it is the min and the max. */
- if (oss_num_threads > OSS_THREADS_MAX)
+ if (oss_num_threads > OSS_THREADS_MAX)
oss_num_threads = OSS_THREADS_MAX;
if (oss_num_threads < OSS_THREADS_MIN)
oss_num_threads = OSS_THREADS_MIN;
oss_max_threads = oss_min_threads = oss_num_threads;
} else {
/* Base min threads on memory and cpus */
- oss_min_threads = num_possible_cpus() * num_physpages >>
+ oss_min_threads = num_possible_cpus() * num_physpages >>
(27 - CFS_PAGE_SHIFT);
if (oss_min_threads < OSS_THREADS_MIN)
oss_min_threads = OSS_THREADS_MIN;
/* Insure a 4x range for dynamic threads */
- if (oss_min_threads > OSS_THREADS_MAX / 4)
+ if (oss_min_threads > OSS_THREADS_MAX / 4)
oss_min_threads = OSS_THREADS_MAX / 4;
oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4);
}
ost->ost_service =
ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
OST_MAXREPSIZE, OST_REQUEST_PORTAL,
- OSC_REPLY_PORTAL,
- OST_WATCHDOG_TIMEOUT, ost_handle,
- LUSTRE_OSS_NAME, obd->obd_proc_entry,
- ost_print_req, oss_min_threads,
- oss_max_threads, "ll_ost",
- LCT_DT_THREAD);
+ OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
+ ost_handle, LUSTRE_OSS_NAME,
+ obd->obd_proc_entry, target_print_req,
+ oss_min_threads, oss_max_threads,
+ "ll_ost", LCT_DT_THREAD);
if (ost->ost_service == NULL) {
CERROR("failed to start service\n");
GOTO(out_lprocfs, rc = -ENOMEM);
if (oss_num_create_threads) {
if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
oss_num_create_threads = OSS_MAX_CREATE_THREADS;
- if (oss_num_create_threads < OSS_DEF_CREATE_THREADS)
- oss_num_create_threads = OSS_DEF_CREATE_THREADS;
+ if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
+ oss_num_create_threads = OSS_MIN_CREATE_THREADS;
oss_min_create_threads = oss_max_create_threads =
oss_num_create_threads;
} else {
- oss_min_create_threads = OSS_DEF_CREATE_THREADS;
+ oss_min_create_threads = OSS_MIN_CREATE_THREADS;
oss_max_create_threads = OSS_MAX_CREATE_THREADS;
}
ost->ost_create_service =
ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
OST_MAXREPSIZE, OST_CREATE_PORTAL,
- OSC_REPLY_PORTAL,
- OST_WATCHDOG_TIMEOUT, ost_handle, "ost_create",
- obd->obd_proc_entry, ost_print_req,
- oss_min_create_threads,
- oss_max_create_threads,
+ OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
+ ost_handle, "ost_create",
+ obd->obd_proc_entry, target_print_req,
+ oss_min_create_threads, oss_max_create_threads,
"ll_ost_creat", LCT_DT_THREAD);
if (ost->ost_create_service == NULL) {
CERROR("failed to start OST create service\n");
ost->ost_io_service =
ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
OST_MAXREPSIZE, OST_IO_PORTAL,
- OSC_REPLY_PORTAL,
- OST_WATCHDOG_TIMEOUT, ost_handle, "ost_io",
- obd->obd_proc_entry, ost_print_req,
+ OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
+ ost_handle, "ost_io",
+ obd->obd_proc_entry, target_print_req,
oss_min_threads, oss_max_threads,
"ll_ost_io", LCT_DT_THREAD);
if (ost->ost_io_service == NULL) {
class_unregister_type(LUSTRE_OSS_NAME);
}
-MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
+MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
MODULE_LICENSE("GPL");