-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
/*
* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*/
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MDC
#ifdef __KERNEL__
int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
__u64 *bits)
{
- struct ldlm_lock *lock;
+ struct ldlm_lock *lock;
+ struct inode *new_inode = data;
ENTRY;
if(bits)
LASSERT(lock != NULL);
lock_res_and_lock(lock);
#ifdef __KERNEL__
- if (lock->l_ast_data && lock->l_ast_data != data) {
- struct inode *new_inode = data;
- struct inode *old_inode = lock->l_ast_data;
- LASSERTF(old_inode->i_state & I_FREEING,
- "Found existing inode %p/%lu/%u state %lu in lock: "
- "setting data to %p/%lu/%u\n", old_inode,
- old_inode->i_ino, old_inode->i_generation,
- old_inode->i_state,
- new_inode, new_inode->i_ino, new_inode->i_generation);
- }
+ if (lock->l_resource->lr_lvb_inode &&
+ lock->l_resource->lr_lvb_inode != data) {
+ struct inode *old_inode = lock->l_resource->lr_lvb_inode;
+ LASSERTF(old_inode->i_state & I_FREEING,
+ "Found existing inode %p/%lu/%u state %lu in lock: "
+ "setting data to %p/%lu/%u\n", old_inode,
+ old_inode->i_ino, old_inode->i_generation,
+ old_inode->i_state,
+ new_inode, new_inode->i_ino, new_inode->i_generation);
+ }
#endif
- lock->l_ast_data = data;
+ lock->l_resource->lr_lvb_inode = new_inode;
if (bits)
*bits = lock->l_policy_data.l_inodebits.bits;
RETURN(0);
}
-ldlm_mode_t mdc_lock_match(struct obd_export *exp, int flags,
+ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
const struct lu_fid *fid, ldlm_type_t type,
ldlm_policy_data_t *policy, ldlm_mode_t mode,
struct lustre_handle *lockh)
RETURN(rc);
}
-int mdc_change_cbdata(struct obd_export *exp,
- const struct lu_fid *fid,
- ldlm_iterator_t it, void *data)
+int mdc_null_inode(struct obd_export *exp,
+ const struct lu_fid *fid)
{
- struct ldlm_res_id res_id;
- ENTRY;
+ struct ldlm_res_id res_id;
+ struct ldlm_resource *res;
+ struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
+ ENTRY;
- fid_build_reg_res_name(fid, &res_id);
- ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace,
- &res_id, it, data);
+ LASSERTF(ns != NULL, "no namespace passed\n");
- EXIT;
- return 0;
+ fid_build_reg_res_name(fid, &res_id);
+
+ res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
+ if(res == NULL)
+ RETURN(0);
+
+ lock_res(res);
+ res->lr_lvb_inode = NULL;
+ unlock_res(res);
+
+ ldlm_resource_putref(res);
+ RETURN(0);
}
/* find any ldlm lock of the inode in mdc
static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
{
- /* Don't hold error requests for replay. */
- if (req->rq_replay) {
- cfs_spin_lock(&req->rq_lock);
- req->rq_replay = 0;
- cfs_spin_unlock(&req->rq_lock);
+ /* Don't hold error requests for replay. */
+ if (req->rq_replay) {
+ spin_lock(&req->rq_lock);
+ req->rq_replay = 0;
+ spin_unlock(&req->rq_lock);
}
if (rc && req->rq_transno != 0) {
DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
return NULL;
}
- cfs_spin_lock(&req->rq_lock);
- req->rq_replay = req->rq_import->imp_replayable;
- cfs_spin_unlock(&req->rq_lock);
+ spin_lock(&req->rq_lock);
+ req->rq_replay = req->rq_import->imp_replayable;
+ spin_unlock(&req->rq_lock);
/* pack the intent */
lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
RETURN(req);
}
-static struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp)
+static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
+ struct lookup_intent *it,
+ struct md_op_data *unused)
+{
+ struct obd_device *obd = class_exp2obd(exp);
+ struct ptlrpc_request *req;
+ struct ldlm_intent *lit;
+ struct layout_intent *layout;
+ int rc;
+ ENTRY;
+
+ req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+ &RQF_LDLM_INTENT_LAYOUT);
+ if (req == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
+
+ req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
+ rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+ if (rc) {
+ ptlrpc_request_free(req);
+ RETURN(ERR_PTR(rc));
+ }
+
+ /* pack the intent */
+ lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+ lit->opc = (__u64)it->it_op;
+
+ /* pack the layout intent request */
+ layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
+ /* LAYOUT_INTENT_ACCESS is generic, specific operation will be
+ * set for replication */
+ layout->li_opc = LAYOUT_INTENT_ACCESS;
+
+ req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
+ obd->u.cli.cl_max_mds_easize);
+ ptlrpc_request_set_replen(req);
+ RETURN(req);
+}
+
+static struct ptlrpc_request *
+mdc_enqueue_pack(struct obd_export *exp, int lvb_len)
{
struct ptlrpc_request *req;
int rc;
RETURN(ERR_PTR(rc));
}
+ req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
ptlrpc_request_set_replen(req);
RETURN(req);
}
struct lustre_handle *lockh,
int rc)
{
- struct req_capsule *pill = &req->rq_pill;
- struct ldlm_request *lockreq;
- struct ldlm_reply *lockrep;
+ struct req_capsule *pill = &req->rq_pill;
+ struct ldlm_request *lockreq;
+ struct ldlm_reply *lockrep;
+ struct lustre_intent_data *intent = &it->d.lustre;
+ struct ldlm_lock *lock;
+ void *lvb_data = NULL;
+ int lvb_len = 0;
ENTRY;
LASSERT(rc >= 0);
* actually get a lock, just perform the intent. */
if (req->rq_transno || req->rq_replay) {
lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
- lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
+ lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
}
if (rc == ELDLM_LOCK_ABORTED) {
memset(lockh, 0, sizeof(*lockh));
rc = 0;
} else { /* rc = 0 */
- struct ldlm_lock *lock = ldlm_handle2lock(lockh);
- LASSERT(lock);
+ lock = ldlm_handle2lock(lockh);
+ LASSERT(lock != NULL);
/* If the server gave us back a different lock mode, we should
* fix up our variables. */
ldlm_lock_decref(lockh, einfo->ei_mode);
einfo->ei_mode = lock->l_req_mode;
}
- LDLM_LOCK_PUT(lock);
- }
+ LDLM_LOCK_PUT(lock);
+ }
- lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
- LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
+ lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
+ LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
- it->d.lustre.it_disposition = (int)lockrep->lock_policy_res1;
- it->d.lustre.it_status = (int)lockrep->lock_policy_res2;
- it->d.lustre.it_lock_mode = einfo->ei_mode;
- it->d.lustre.it_lock_handle = lockh->cookie;
- it->d.lustre.it_data = req;
+ intent->it_disposition = (int)lockrep->lock_policy_res1;
+ intent->it_status = (int)lockrep->lock_policy_res2;
+ intent->it_lock_mode = einfo->ei_mode;
+ intent->it_lock_handle = lockh->cookie;
+ intent->it_data = req;
- if (it->d.lustre.it_status < 0 && req->rq_replay)
- mdc_clear_replay_flag(req, it->d.lustre.it_status);
+ /* Technically speaking rq_transno must already be zero if
+ * it_status is in error, so the check is a bit redundant */
+ if ((!req->rq_transno || intent->it_status < 0) && req->rq_replay)
+ mdc_clear_replay_flag(req, intent->it_status);
/* If we're doing an IT_OPEN which did not result in an actual
* successful open, then we need to remove the bit which saves
* function without doing so, and try to replay a failed create
* (bug 3440) */
if (it->it_op & IT_OPEN && req->rq_replay &&
- (!it_disposition(it, DISP_OPEN_OPEN) ||it->d.lustre.it_status != 0))
- mdc_clear_replay_flag(req, it->d.lustre.it_status);
+ (!it_disposition(it, DISP_OPEN_OPEN) ||intent->it_status != 0))
+ mdc_clear_replay_flag(req, intent->it_status);
- DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
- it->it_op,it->d.lustre.it_disposition,it->d.lustre.it_status);
+ DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
+ it->it_op, intent->it_disposition, intent->it_status);
/* We know what to expect, so we do any byte flipping required here */
if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
* is swabbed by that handler correctly.
*/
mdc_set_open_replay_data(NULL, NULL, req);
- }
+ }
if ((body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) != 0) {
void *eadata;
- mdc_update_max_ea_from_body(exp, body);
+ mdc_update_max_ea_from_body(exp, body);
/*
* The eadata is opaque; just check that it is there.
if (eadata == NULL)
RETURN(-EPROTO);
+ /* save lvb data and length in case this is for layout
+ * lock */
+ lvb_data = eadata;
+ lvb_len = body->eadatasize;
+
/*
* We save the reply LOV EA in case we have to replay a
* create for recovery. If we didn't allocate a large
if (capa == NULL)
RETURN(-EPROTO);
}
- }
-
- RETURN(rc);
+ } else if (it->it_op & IT_LAYOUT) {
+ /* maybe the lock was granted right away and layout
+ * is packed into RMF_DLM_LVB of req */
+ lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
+ if (lvb_len > 0) {
+ lvb_data = req_capsule_server_sized_get(pill,
+ &RMF_DLM_LVB, lvb_len);
+ if (lvb_data == NULL)
+ RETURN(-EPROTO);
+ }
+ }
+
+ /* fill in stripe data for layout lock */
+ lock = ldlm_handle2lock(lockh);
+ if (lock != NULL && ldlm_has_layout(lock) && lvb_data != NULL) {
+ void *lmm;
+
+ LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d\n",
+ ldlm_it2str(it->it_op), lvb_len);
+
+ OBD_ALLOC_LARGE(lmm, lvb_len);
+ if (lmm == NULL) {
+ LDLM_LOCK_PUT(lock);
+ RETURN(-ENOMEM);
+ }
+ memcpy(lmm, lvb_data, lvb_len);
+
+ /* install lvb_data */
+ lock_res_and_lock(lock);
+ if (lock->l_lvb_data == NULL) {
+ lock->l_lvb_data = lmm;
+ lock->l_lvb_len = lvb_len;
+ lmm = NULL;
+ }
+ unlock_res_and_lock(lock);
+ if (lmm != NULL)
+ OBD_FREE_LARGE(lmm, lvb_len);
+ }
+ if (lock != NULL)
+ LDLM_LOCK_PUT(lock);
+
+ RETURN(rc);
}
/* We always reserve enough space in the reply packet for a stripe MD, because
int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
struct lookup_intent *it, struct md_op_data *op_data,
struct lustre_handle *lockh, void *lmm, int lmmsize,
- struct ptlrpc_request **reqp, int extra_lock_flags)
+ struct ptlrpc_request **reqp, __u64 extra_lock_flags)
{
struct obd_device *obddev = class_exp2obd(exp);
struct ptlrpc_request *req = NULL;
- struct req_capsule *pill;
- int flags = extra_lock_flags;
+ __u64 flags, saved_flags = extra_lock_flags;
int rc;
struct ldlm_res_id res_id;
static const ldlm_policy_data_t lookup_policy =
{ .l_inodebits = { MDS_INODELOCK_LOOKUP } };
static const ldlm_policy_data_t update_policy =
{ .l_inodebits = { MDS_INODELOCK_UPDATE } };
+ static const ldlm_policy_data_t layout_policy =
+ { .l_inodebits = { MDS_INODELOCK_LAYOUT } };
ldlm_policy_data_t const *policy = &lookup_policy;
+ int generation, resends = 0;
+ struct ldlm_reply *lockrep;
+ enum lvb_type lvb_type = 0;
ENTRY;
LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
fid_build_reg_res_name(&op_data->op_fid1, &res_id);
- if (it)
- flags |= LDLM_FL_HAS_INTENT;
- if (it && it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
- policy = &update_policy;
+ if (it) {
+ saved_flags |= LDLM_FL_HAS_INTENT;
+ if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
+ policy = &update_policy;
+ else if (it->it_op & IT_LAYOUT)
+ policy = &layout_policy;
+ }
- if (reqp)
- req = *reqp;
+ LASSERT(reqp == NULL);
+ generation = obddev->u.cli.cl_import->imp_generation;
+resend:
+ flags = saved_flags;
if (!it) {
/* The only way right now is FLOCK, in this case we hide flock
policy as lmm, but lmmsize is 0 */
policy = &update_policy;
einfo->ei_cbdata = NULL;
lmm = NULL;
- } else if (it->it_op & IT_UNLINK)
- req = mdc_intent_unlink_pack(exp, it, op_data);
- else if (it->it_op & (IT_GETATTR | IT_LOOKUP))
- req = mdc_intent_getattr_pack(exp, it, op_data);
- else if (it->it_op == IT_READDIR)
- req = ldlm_enqueue_pack(exp);
- else {
+ } else if (it->it_op & IT_UNLINK) {
+ req = mdc_intent_unlink_pack(exp, it, op_data);
+ } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
+ req = mdc_intent_getattr_pack(exp, it, op_data);
+ } else if (it->it_op & IT_READDIR) {
+ req = mdc_enqueue_pack(exp, 0);
+ } else if (it->it_op & IT_LAYOUT) {
+ if (!imp_connect_lvb_type(class_exp2cliimp(exp)))
+ RETURN(-EOPNOTSUPP);
+
+ req = mdc_intent_layout_pack(exp, it, op_data);
+ lvb_type = LVB_T_LAYOUT;
+ } else {
LBUG();
RETURN(-EINVAL);
}
if (IS_ERR(req))
RETURN(PTR_ERR(req));
- pill = &req->rq_pill;
+
+ if (req != NULL && it && it->it_op & IT_CREAT)
+ /* ask ptlrpc not to resend on EINPROGRESS since we have our own
+ * retry logic */
+ req->rq_no_retry_einprogress = 1;
+
+ if (resends) {
+ req->rq_generation_set = 1;
+ req->rq_import_generation = generation;
+ req->rq_sent = cfs_time_current_sec() + resends;
+ }
/* It is important to obtain rpc_lock first (if applicable), so that
* threads that are serialised with rpc_lock are not polluting our
rc = mdc_enter_request(&obddev->u.cli);
if (rc != 0) {
mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+ mdc_clear_replay_flag(req, 0);
ptlrpc_req_finished(req);
RETURN(rc);
}
}
rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
- 0, lockh, 0);
- if (reqp)
- *reqp = req;
-
- if (it) {
- mdc_exit_request(&obddev->u.cli);
- mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
- }
+ 0, lvb_type, lockh, 0);
if (!it) {
/* For flock requests we immediatelly return without further
delay and let caller deal with the rest, since rest of
RETURN(rc);
}
+ mdc_exit_request(&obddev->u.cli);
+ mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
+
if (rc < 0) {
CERROR("ldlm_cli_enqueue: %d\n", rc);
mdc_clear_replay_flag(req, rc);
ptlrpc_req_finished(req);
RETURN(rc);
}
- rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
- RETURN(rc);
+ lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+ LASSERT(lockrep != NULL);
+
+ lockrep->lock_policy_res2 =
+ ptlrpc_status_ntoh(lockrep->lock_policy_res2);
+
+ /* Retry the create infinitely when we get -EINPROGRESS from
+ * server. This is required by the new quota design. */
+ if (it && it->it_op & IT_CREAT &&
+ (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
+ mdc_clear_replay_flag(req, rc);
+ ptlrpc_req_finished(req);
+ resends++;
+
+ CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
+ obddev->obd_name, resends, it->it_op,
+ PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
+
+ if (generation == obddev->u.cli.cl_import->imp_generation) {
+ goto resend;
+ } else {
+ CDEBUG(D_HA, "resend cross eviction\n");
+ RETURN(-EIO);
+ }
+ }
+
+ rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
+ if (rc < 0) {
+ if (lustre_handle_is_used(lockh)) {
+ ldlm_lock_decref(lockh, einfo->ei_mode);
+ memset(lockh, 0, sizeof(*lockh));
+ }
+ ptlrpc_req_finished(req);
+ }
+ RETURN(rc);
}
static int mdc_finish_intent_lock(struct obd_export *exp,
struct mdt_body *mdt_body;
struct ldlm_lock *lock;
int rc;
-
+ ENTRY;
LASSERT(request != NULL);
LASSERT(request != LP_POISON);
} else if (it->it_op == IT_OPEN) {
LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
} else {
- LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
+ LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_LAYOUT));
}
/* If we already have a matching lock, then cancel the new
mode = ldlm_revalidate_lock_handle(&lockh, bits);
} else {
fid_build_reg_res_name(fid, &res_id);
- policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
- MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
+ switch (it->it_op) {
+ case IT_GETATTR:
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
+ break;
+ case IT_LAYOUT:
+ policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
+ break;
+ default:
+ policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
+ break;
+ }
mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
LDLM_FL_BLOCK_GRANTED, &res_id,
LDLM_IBITS, &policy,
void *lmm, int lmmsize, struct lookup_intent *it,
int lookup_flags, struct ptlrpc_request **reqp,
ldlm_blocking_callback cb_blocking,
- int extra_lock_flags)
+ __u64 extra_lock_flags)
{
struct lustre_handle lockh;
int rc = 0;
struct lookup_intent *it;
struct lustre_handle *lockh;
struct obd_device *obddev;
- int flags = LDLM_FL_HAS_INTENT;
+ struct ldlm_reply *lockrep;
+ __u64 flags = LDLM_FL_HAS_INTENT;
ENTRY;
it = &minfo->mi_it;
GOTO(out, rc);
}
+ lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+ LASSERT(lockrep != NULL);
+
+ lockrep->lock_policy_res2 =
+ ptlrpc_status_ntoh(lockrep->lock_policy_res2);
+
rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
if (rc)
GOTO(out, rc);
MDS_INODELOCK_UPDATE }
};
int rc = 0;
- int flags = LDLM_FL_HAS_INTENT;
+ __u64 flags = LDLM_FL_HAS_INTENT;
ENTRY;
CDEBUG(D_DLMTRACE,"name: %.*s in inode "DFID", intent: %s flags %#o\n",
}
rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
- 0, &minfo->mi_lockh, 1);
+ 0, LVB_T_NONE, &minfo->mi_lockh, 1);
if (rc < 0) {
mdc_exit_request(&obddev->u.cli);
ptlrpc_req_finished(req);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff