-/* -*- 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.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * 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__
#include <lprocfs_status.h>
#include "mdc_internal.h"
+struct mdc_getattr_args {
+ struct obd_export *ga_exp;
+ struct md_enqueue_info *ga_minfo;
+ struct ldlm_enqueue_info *ga_einfo;
+};
+
int it_disposition(struct lookup_intent *it, int flag)
{
return it->d.lustre.it_disposition & flag;
int it_open_error(int phase, struct lookup_intent *it)
{
+ if (it_disposition(it, DISP_OPEN_LEASE)) {
+ if (phase >= DISP_OPEN_LEASE)
+ return it->d.lustre.it_status;
+ else
+ return 0;
+ }
if (it_disposition(it, DISP_OPEN_OPEN)) {
if (phase >= DISP_OPEN_OPEN)
return it->d.lustre.it_status;
EXPORT_SYMBOL(it_open_error);
/* this must be called on a lockh that is known to have a referenced lock */
-int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
+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 (!*lockh) {
- EXIT;
+ if(bits)
+ *bits = 0;
+
+ if (!*lockh)
RETURN(0);
- }
lock = ldlm_handle2lock((struct lustre_handle *)lockh);
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;
+
unlock_res_and_lock(lock);
LDLM_LOCK_PUT(lock);
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)
int mdc_cancel_unused(struct obd_export *exp,
const struct lu_fid *fid,
ldlm_policy_data_t *policy,
- ldlm_mode_t mode, int flags, void *opaque)
+ ldlm_mode_t mode,
+ ldlm_cancel_flags_t flags,
+ void *opaque)
{
struct ldlm_res_id res_id;
struct obd_device *obd = class_exp2obd(exp);
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;
+ struct ldlm_resource *res;
+ struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
+ ENTRY;
+
+ LASSERTF(ns != NULL, "no namespace passed\n");
+
+ 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
+ * return 0 not find
+ * 1 find one
+ * < 0 error */
+int mdc_find_cbdata(struct obd_export *exp,
+ const struct lu_fid *fid,
+ ldlm_iterator_t it, void *data)
{
struct ldlm_res_id res_id;
+ int rc = 0;
ENTRY;
- fid_build_reg_res_name(fid, &res_id);
- ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace,
- &res_id, it, data);
-
- EXIT;
- return 0;
+ fid_build_reg_res_name((struct lu_fid*)fid, &res_id);
+ rc = ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace, &res_id,
+ it, data);
+ if (rc == LDLM_ITER_STOP)
+ RETURN(1);
+ else if (rc == LDLM_ITER_CONTINUE)
+ RETURN(0);
+ RETURN(rc);
}
static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
{
- /* 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);
+ /* 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);
struct ptlrpc_request *req;
struct obd_device *obddev = class_exp2obd(exp);
struct ldlm_intent *lit;
- int joinfile = !!((it->it_flags & O_JOIN_FILE) &&
- op_data->op_data);
CFS_LIST_HEAD(cancels);
int count = 0;
int mode;
/* XXX: openlock is not cancelled for cross-refs. */
/* If inode is known, cancel conflicting OPEN locks. */
- if (fid_is_sane(&op_data->op_fid2)) {
- if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
- mode = LCK_CW;
+ if (fid_is_sane(&op_data->op_fid2)) {
+ if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
+ if (it->it_flags & FMODE_WRITE)
+ mode = LCK_EX;
+ else
+ mode = LCK_PR;
+ } else {
+ if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
+ mode = LCK_CW;
#ifdef FMODE_EXEC
- else if (it->it_flags & FMODE_EXEC)
- mode = LCK_PR;
+ else if (it->it_flags & FMODE_EXEC)
+ mode = LCK_PR;
#endif
- else
- mode = LCK_CR;
- count = mdc_resource_get_unused(exp, &op_data->op_fid2,
- &cancels, mode,
- MDS_INODELOCK_OPEN);
- }
-
- /* If CREATE or JOIN_FILE, cancel parent's UPDATE lock. */
- if (it->it_op & IT_CREAT || joinfile)
+ else
+ mode = LCK_CR;
+ }
+ count = mdc_resource_get_unused(exp, &op_data->op_fid2,
+ &cancels, mode,
+ MDS_INODELOCK_OPEN);
+ }
+
+ /* If CREATE, cancel parent's UPDATE lock. */
+ if (it->it_op & IT_CREAT)
mode = LCK_EX;
else
mode = LCK_CR;
op_data->op_namelen + 1);
req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
max(lmmsize, obddev->u.cli.cl_default_mds_easize));
- if (!joinfile) {
- req_capsule_set_size(&req->rq_pill, &RMF_REC_JOINFILE,
- RCL_CLIENT, 0);
- }
rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
if (rc) {
return NULL;
}
- if (joinfile) {
- __u64 head_size = *(__u64 *)op_data->op_data;
- mdc_join_pack(req, op_data, head_size);
- }
-
- spin_lock(&req->rq_lock);
- req->rq_replay = 1;
- 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 *
+mdc_intent_getxattr_pack(struct obd_export *exp,
+ struct lookup_intent *it,
+ struct md_op_data *op_data)
+{
+ struct ptlrpc_request *req;
+ struct ldlm_intent *lit;
+ int rc, count = 0, maxdata;
+ CFS_LIST_HEAD(cancels);
+
+ ENTRY;
+
+ req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+ &RQF_LDLM_INTENT_GETXATTR);
+ if (req == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
+
+ mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+
+ rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
+ 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 = IT_GETXATTR;
+
+ maxdata = class_exp2cliimp(exp)->imp_connect_data.ocd_max_easize;
+
+ /* pack the intended request */
+ mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
+ op_data->op_valid, maxdata, -1, 0);
+
+ req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
+ RCL_SERVER, maxdata);
+
+ req_capsule_set_size(&req->rq_pill, &RMF_EAVALS,
+ RCL_SERVER, maxdata);
+
+ req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS,
+ RCL_SERVER, maxdata);
+
+ ptlrpc_request_set_replen(req);
+
+ RETURN(req);
+}
+
static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
struct lookup_intent *it,
struct md_op_data *op_data)
lit->opc = (__u64)it->it_op;
/* pack the intended request */
- mdc_getattr_pack(req, valid, it->it_flags, op_data);
+ mdc_getattr_pack(req, valid, it->it_flags, op_data,
+ obddev->u.cli.cl_max_mds_easize);
req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
obddev->u.cli.cl_max_mds_easize);
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_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);
+
/*
* The eadata is opaque; just check that it is there.
* Eventually, obd_unpackmd() will check the contents.
if (eadata == NULL)
RETURN(-EPROTO);
- if (body->valid & OBD_MD_FLMODEASIZE) {
- struct obd_device *obddev = class_exp2obd(exp);
-
- if (obddev->u.cli.cl_max_mds_easize <
- body->max_mdsize) {
- obddev->u.cli.cl_max_mds_easize =
- body->max_mdsize;
- CDEBUG(D_INFO, "maxeasize become %d\n",
- body->max_mdsize);
- }
- if (obddev->u.cli.cl_max_mds_cookiesize <
- body->max_cookiesize) {
- obddev->u.cli.cl_max_mds_cookiesize =
- body->max_cookiesize;
- CDEBUG(D_INFO, "cookiesize become %d\n",
- body->max_cookiesize);
- }
- }
+ /* 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
void *lmm;
if (req_capsule_get_size(pill, &RMF_EADATA,
RCL_CLIENT) <
- body->eadatasize) {
+ body->eadatasize)
mdc_realloc_openmsg(req, body);
- req_capsule_set_size(pill, &RMF_EADATA,
- RCL_CLIENT,
- body->eadatasize);
- }
+ else
+ req_capsule_shrink(pill, &RMF_EADATA,
+ body->eadatasize,
+ RCL_CLIENT);
+
+ req_capsule_set_size(pill, &RMF_EADATA,
+ RCL_CLIENT,
+ body->eadatasize);
+
lmm = req_capsule_client_get(pill, &RMF_EADATA);
if (lmm)
memcpy(lmm, eadata, body->eadatasize);
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;
- ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
+ 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 } };
+ static const ldlm_policy_data_t getxattr_policy = {
+ .l_inodebits = { MDS_INODELOCK_XATTR } };
+ 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.l_inodebits.bits = MDS_INODELOCK_UPDATE;
-
- if (reqp)
- req = *reqp;
-
+ 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;
+ else if (it->it_op & (IT_GETXATTR | IT_SETXATTR))
+ policy = &getxattr_policy;
+ }
+
+ 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 */
LASSERT(lmm && lmmsize == 0);
LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
einfo->ei_type);
- policy = *(ldlm_policy_data_t *)lmm;
+ policy = (ldlm_policy_data_t *)lmm;
res_id.name[3] = LDLM_FLOCK;
} else if (it->it_op & IT_OPEN) {
- int joinfile = !!((it->it_flags & O_JOIN_FILE) &&
- op_data->op_data);
-
req = mdc_intent_open_pack(exp, it, op_data, lmm, lmmsize,
einfo->ei_cbdata);
- if (!joinfile) {
- policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
- einfo->ei_cbdata = NULL;
- lmm = NULL;
- } else
- it->it_flags &= ~O_JOIN_FILE;
- } 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 {
+ 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 = 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 if (it->it_op & IT_GETXATTR) {
+ req = mdc_intent_getxattr_pack(exp, it, op_data);
+ } 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
* rpcs in flight counter. We do not do flock request limiting, though*/
if (it) {
mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
- mdc_enter_request(&obddev->u.cli);
+ 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, NULL, 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);
- }
+ rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
+ 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
this function metadata processing makes no sense for flock
- requests anyway */
+ requests anyway. But in case of problem during comms with
+ Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
+ can not rely on caller and this mainly for F_UNLCKs
+ (explicits or automatically generated by Kernel to clean
+ current FLocks upon exit) that can't be trashed */
+ if (((rc == -EINTR) || (rc == -ETIMEDOUT)) &&
+ (einfo->ei_type == LDLM_FLOCK) &&
+ (einfo->ei_mode == LCK_NL))
+ goto resend;
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);
/* If we were revalidating a fid/name pair, mark the intent in
* case we fail and get called again from lookup */
if (fid_is_sane(&op_data->op_fid2) &&
- it->it_flags & O_CHECK_STALE &&
+ it->it_create_mode & M_CHECK_STALE &&
it->it_op != IT_GETATTR) {
it_set_disposition(it, DISP_ENQ_COMPLETE);
} 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
ldlm_policy_data_t policy = lock->l_policy_data;
LDLM_DEBUG(lock, "matching against this");
- LASSERTF(fid_res_name_eq(&mdt_body->fid1,
- &lock->l_resource->lr_name),
- "Lock res_id: %lu/%lu/%lu, fid: %lu/%lu/%lu.\n",
- (unsigned long)lock->l_resource->lr_name.name[0],
- (unsigned long)lock->l_resource->lr_name.name[1],
- (unsigned long)lock->l_resource->lr_name.name[2],
- (unsigned long)fid_seq(&mdt_body->fid1),
- (unsigned long)fid_oid(&mdt_body->fid1),
- (unsigned long)fid_ver(&mdt_body->fid1));
- LDLM_LOCK_PUT(lock);
+ LASSERTF(fid_res_name_eq(&mdt_body->fid1,
+ &lock->l_resource->lr_name),
+ "Lock res_id: "DLDLMRES", fid: "DFID"\n",
+ PLDLMRES(lock->l_resource), PFID(&mdt_body->fid1));
+ LDLM_LOCK_PUT(lock);
memcpy(&old_lock, lockh, sizeof(*lockh));
if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
RETURN(rc);
}
+int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
+ struct lu_fid *fid, __u64 *bits)
+{
+ /* We could just return 1 immediately, but since we should only
+ * be called in revalidate_it if we already have a lock, let's
+ * verify that. */
+ struct ldlm_res_id res_id;
+ struct lustre_handle lockh;
+ ldlm_policy_data_t policy;
+ ldlm_mode_t mode;
+ ENTRY;
+
+ if (it->d.lustre.it_lock_handle) {
+ lockh.cookie = it->d.lustre.it_lock_handle;
+ mode = ldlm_revalidate_lock_handle(&lockh, bits);
+ } else {
+ fid_build_reg_res_name(fid, &res_id);
+ switch (it->it_op) {
+ case IT_GETATTR:
+ /* File attributes are held under multiple bits:
+ * nlink is under lookup lock, size and times are
+ * under UPDATE lock and recently we've also got
+ * a separate permissions lock for owner/group/acl that
+ * were protected by lookup lock before.
+ * Getattr must provide all of that information,
+ * so we need to ensure we have all of those locks.
+ * Unfortunately, if the bits are split across multiple
+ * locks, there's no easy way to match all of them here,
+ * so an extra RPC would be performed to fetch all
+ * of those bits at once for now. */
+ policy.l_inodebits.bits = MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_LOOKUP |
+ MDS_INODELOCK_PERM;
+ 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,
+ LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh, 0);
+ }
+
+ if (mode) {
+ it->d.lustre.it_lock_handle = lockh.cookie;
+ it->d.lustre.it_lock_mode = mode;
+ } else {
+ it->d.lustre.it_lock_handle = 0;
+ it->d.lustre.it_lock_mode = 0;
+ }
+
+ RETURN(!!mode);
+}
+
/*
* This long block is all about fixing up the lock and request state
* so that it is correct as of the moment _before_ the operation was
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;
- ENTRY;
- LASSERT(it);
+ struct lustre_handle lockh;
+ int rc = 0;
+ ENTRY;
+ LASSERT(it);
- CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
- ", intent: %s flags %#o\n", op_data->op_namelen,
- op_data->op_name, PFID(&op_data->op_fid2),
- PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
- it->it_flags);
+ CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
+ ", intent: %s flags %#Lo\n", op_data->op_namelen,
+ op_data->op_name, PFID(&op_data->op_fid2),
+ PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
+ it->it_flags);
lockh.cookie = 0;
if (fid_is_sane(&op_data->op_fid2) &&
/* We could just return 1 immediately, but since we should only
* be called in revalidate_it if we already have a lock, let's
* verify that. */
- ldlm_policy_data_t policy;
- ldlm_mode_t mode;
-
- /* As not all attributes are kept under update lock, e.g.
- owner/group/acls are under lookup lock, we need both
- ibits for GETATTR. */
-
- /* For CMD, UPDATE lock and LOOKUP lock can not be got
- * at the same for cross-object, so we can not match
- * the 2 lock at the same time FIXME: but how to handle
- * the above situation */
- policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
- MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
-
- mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED,
- &op_data->op_fid2, LDLM_IBITS, &policy,
- LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh);
- if (mode) {
- it->d.lustre.it_lock_handle = lockh.cookie;
- it->d.lustre.it_lock_mode = mode;
- }
-
+ it->d.lustre.it_lock_handle = 0;
+ rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
/* Only return failure if it was not GETATTR by cfid
(from inode_revalidate) */
- if (mode || op_data->op_namelen != 0)
- RETURN(!!mode);
+ if (rc || op_data->op_namelen != 0)
+ RETURN(rc);
}
/* lookup_it may be called only after revalidate_it has run, because
* this and use the request from revalidate. In this case, revalidate
* never dropped its reference, so the refcounts are all OK */
if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
- struct ldlm_enqueue_info einfo =
- { LDLM_IBITS, it_to_lock_mode(it), cb_blocking,
- ldlm_completion_ast, NULL, NULL, NULL };
+ struct ldlm_enqueue_info einfo = {
+ .ei_type = LDLM_IBITS,
+ .ei_mode = it_to_lock_mode(it),
+ .ei_cb_bl = cb_blocking,
+ .ei_cb_cp = ldlm_completion_ast,
+ };
/* For case if upper layer did not alloc fid, do it now. */
if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
lmm, lmmsize, NULL, extra_lock_flags);
if (rc < 0)
RETURN(rc);
- it->d.lustre.it_lock_handle = lockh.cookie;
} else if (!fid_is_sane(&op_data->op_fid2) ||
- !(it->it_flags & O_CHECK_STALE)) {
+ !(it->it_create_mode & M_CHECK_STALE)) {
/* DISP_ENQ_COMPLETE set means there is extra reference on
* request referenced from this intent, saved for subsequent
* lookup. This path is executed when we proceed to this
static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
struct ptlrpc_request *req,
- void *unused, int rc)
+ void *args, int rc)
{
- struct obd_export *exp = req->rq_async_args.pointer_arg[0];
- struct md_enqueue_info *minfo = req->rq_async_args.pointer_arg[1];
- struct ldlm_enqueue_info *einfo = req->rq_async_args.pointer_arg[2];
+ struct mdc_getattr_args *ga = args;
+ struct obd_export *exp = ga->ga_exp;
+ struct md_enqueue_info *minfo = ga->ga_minfo;
+ struct ldlm_enqueue_info *einfo = ga->ga_einfo;
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;
rc = -ETIMEDOUT;
rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
- &flags, NULL, 0, NULL, lockh, rc);
+ &flags, NULL, 0, lockh, rc);
if (rc < 0) {
CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
mdc_clear_replay_flag(req, rc);
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);
- it->d.lustre.it_lock_handle = lockh->cookie;
-
rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
EXIT;
struct md_op_data *op_data = &minfo->mi_data;
struct lookup_intent *it = &minfo->mi_it;
struct ptlrpc_request *req;
+ struct mdc_getattr_args *ga;
struct obd_device *obddev = class_exp2obd(exp);
struct ldlm_res_id res_id;
+ /*XXX: Both MDS_INODELOCK_LOOKUP and MDS_INODELOCK_UPDATE are needed
+ * for statahead currently. Consider CMD in future, such two bits
+ * maybe managed by different MDS, should be adjusted then. */
ldlm_policy_data_t policy = {
- .l_inodebits = { MDS_INODELOCK_LOOKUP }
+ .l_inodebits = { MDS_INODELOCK_LOOKUP |
+ MDS_INODELOCK_UPDATE }
};
- int rc;
- int flags = LDLM_FL_HAS_INTENT;
- ENTRY;
+ int rc = 0;
+ __u64 flags = LDLM_FL_HAS_INTENT;
+ ENTRY;
- CDEBUG(D_DLMTRACE,"name: %.*s in inode "DFID", intent: %s flags %#o\n",
- op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
- ldlm_it2str(it->it_op), it->it_flags);
+ CDEBUG(D_DLMTRACE,"name: %.*s in inode "DFID", intent: %s flags %#Lo\n",
+ op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
+ ldlm_it2str(it->it_op), it->it_flags);
fid_build_reg_res_name(&op_data->op_fid1, &res_id);
req = mdc_intent_getattr_pack(exp, it, op_data);
if (!req)
RETURN(-ENOMEM);
- mdc_enter_request(&obddev->u.cli);
+ rc = mdc_enter_request(&obddev->u.cli);
+ if (rc != 0) {
+ ptlrpc_req_finished(req);
+ RETURN(rc);
+ }
+
rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
- 0, NULL, &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);
RETURN(rc);
}
- req->rq_async_args.pointer_arg[0] = exp;
- req->rq_async_args.pointer_arg[1] = minfo;
- req->rq_async_args.pointer_arg[2] = einfo;
+ CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
+ ga = ptlrpc_req_async_args(req);
+ ga->ga_exp = exp;
+ ga->ga_minfo = minfo;
+ ga->ga_einfo = einfo;
+
req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
- ptlrpcd_add_req(req, PSCOPE_OTHER);
+ ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
RETURN(0);
}
-
-int mdc_revalidate_lock(struct obd_export *exp,
- struct lookup_intent *it,
- struct lu_fid *fid)
-{
- /* We could just return 1 immediately, but since we should only
- * be called in revalidate_it if we already have a lock, let's
- * verify that. */
- struct ldlm_res_id res_id;
- struct lustre_handle lockh;
- ldlm_policy_data_t policy;
- ldlm_mode_t mode;
- ENTRY;
-
- fid_build_reg_res_name(fid, &res_id);
- /* As not all attributes are kept under update lock, e.g.
- owner/group/acls are under lookup lock, we need both
- ibits for GETATTR. */
- policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
- MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP :
- MDS_INODELOCK_LOOKUP;
-
- mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
- LDLM_FL_BLOCK_GRANTED, &res_id, LDLM_IBITS,
- &policy, LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh, 0);
- if (mode) {
- it->d.lustre.it_lock_handle = lockh.cookie;
- it->d.lustre.it_lock_mode = mode;
- }
-
- RETURN(!!mode);
-}