*
* Copyright (C) 2001-2003 Cluster File Systems, Inc.
*
- * This file is part of Lustre, http://www.sf.net/projects/lustre/
+ * This file is part of the Lustre file system, http://www.lustre.org
+ * Lustre is a trademark of Cluster File Systems, Inc.
*
- * Lustre is free 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.
+ * 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.
*
- * 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
- * GNU General Public License for more details.
+ * 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.
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * 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.
*/
-#define EXPORT_SYMTAB
+#ifndef EXPORT_SYMTAB
+# define EXPORT_SYMTAB
+#endif
#define DEBUG_SUBSYSTEM S_MDC
#ifdef __KERNEL__
# include <linux/init.h>
#else
# include <liblustre.h>
-# include <linux/obd_class.h>
#endif
-#include <linux/lustre_mds.h>
-#include <linux/lustre_lite.h>
-#include <linux/lustre_dlm.h>
-#include <linux/lprocfs_status.h>
+#include <linux/lustre_acl.h>
+#include <obd_class.h>
+#include <lustre_dlm.h>
+#include <lustre_fid.h>
+#include <md_object.h>
+#include <lprocfs_status.h>
+#include <lustre_param.h>
#include "mdc_internal.h"
+static quota_interface_t *quota_interface;
+
#define REQUEST_MINOR 244
-extern int mds_queue_req(struct ptlrpc_request *);
-struct mdc_rpc_lock mdc_rpc_lock;
-struct mdc_rpc_lock mdc_setattr_lock;
-EXPORT_SYMBOL(mdc_rpc_lock);
+static quota_interface_t *quota_interface;
+extern quota_interface_t mdc_quota_interface;
+
+static int mdc_cleanup(struct obd_device *obd);
+
+static struct obd_capa *mdc_unpack_capa(struct ptlrpc_request *req,
+ unsigned int offset)
+{
+ struct lustre_capa *capa;
+ struct obd_capa *oc;
+
+ /* swabbed already in mdc_enqueue */
+ capa = lustre_msg_buf(req->rq_repmsg, offset, sizeof(*capa));
+ if (capa == NULL) {
+ CERROR("missing capa at offset %d failed!\n", offset);
+ return ERR_PTR(-EFAULT);
+ }
+
+ oc = alloc_capa(CAPA_SITE_CLIENT);
+ if (!oc) {
+ CERROR("alloc capa failed!\n");
+ return ERR_PTR(-ENOMEM);
+ }
+ oc->c_capa = *capa;
+
+ return oc;
+}
/* Helper that implements most of mdc_getstatus and signal_completed_replay. */
/* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
-static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
- int level, int msg_flags)
+static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
+ struct obd_capa **pc, int level, int msg_flags)
{
struct ptlrpc_request *req;
- struct mds_body *body;
- int rc, size = sizeof(*body);
+ struct mdt_body *body;
+ int rc, size[3] = { sizeof(struct ptlrpc_body),
+ sizeof(*body),
+ sizeof(struct lustre_capa) };
ENTRY;
- req = ptlrpc_prep_req(imp, MDS_GETSTATUS, 1, &size, NULL);
+ req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS, 2, size,
+ NULL);
if (!req)
GOTO(out, rc = -ENOMEM);
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
- req->rq_level = level;
- req->rq_replen = lustre_msg_size(1, &size);
+ req->rq_send_state = level;
+ ptlrpc_req_set_repsize(req, 3, size);
- mdc_pack_req_body(req);
- req->rq_reqmsg->flags |= msg_flags;
+ mdc_pack_req_body(req, REQ_REC_OFF, 0, NULL, NULL, 0, 0);
+ lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
rc = ptlrpc_queue_wait(req);
if (!rc) {
- body = lustre_swab_repbuf (req, 0, sizeof (*body),
- lustre_swab_mds_body);
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
if (body == NULL) {
- CERROR ("Can't extract mds_body\n");
+ CERROR ("Can't extract mdt_body\n");
GOTO (out, rc = -EPROTO);
}
- memcpy(rootfid, &body->fid1, sizeof(*rootfid));
+ *rootfid = body->fid1;
+
+ if (body->valid & OBD_MD_FLMDSCAPA) {
+ struct obd_capa *oc;
- CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
+ oc = mdc_unpack_capa(req, REPLY_REC_OFF + 1);
+ if (IS_ERR(oc))
+ GOTO(out, rc = PTR_ERR(oc));
+ *pc = oc;
+ }
+
+ CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64
", last_xid="LPU64"\n",
- rootfid->id, req->rq_repmsg->last_committed,
- req->rq_repmsg->last_xid);
+ PFID(rootfid),
+ lustre_msg_get_last_committed(req->rq_repmsg),
+ lustre_msg_get_last_xid(req->rq_repmsg));
}
EXIT;
}
/* This should be mdc_get_info("rootfid") */
-int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid)
-{
- return send_getstatus(class_conn2cliimp(conn), rootfid, LUSTRE_CONN_CON,
- 0);
-}
-
-/* should call mdc_get_info("lovdesc") and mdc_get_info("lovtgts") */
-int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
- struct ptlrpc_request **request)
+int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
+ struct obd_capa **pc)
{
- struct ptlrpc_request *req;
- struct mds_status_req *streq;
- struct lov_desc *desc;
- struct obd_uuid *uuids;
- int rc, size[2] = {sizeof(*streq)};
- int i;
- ENTRY;
-
- req = ptlrpc_prep_req(class_conn2cliimp(mdc_connh), MDS_GETLOVINFO, 1,
- size, NULL);
- if (!req)
- RETURN (-ENOMEM);
-
- *request = req;
- streq = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*streq));
- streq->flags = MDS_STATUS_LOV;
- streq->repbuf = LOV_MAX_UUID_BUFFER_SIZE;
-
- /* prepare for reply */
- req->rq_level = LUSTRE_CONN_CON;
- size[0] = sizeof (*desc);
- size[1] = LOV_MAX_UUID_BUFFER_SIZE;
- req->rq_replen = lustre_msg_size(2, size);
-
- mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
- rc = ptlrpc_queue_wait(req);
- mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
-
- if (rc != 0) {
- CERROR ("rcp failed\n");
- GOTO (failed, rc);
- }
-
- desc = lustre_swab_repbuf (req, 0, sizeof (*desc),
- lustre_swab_lov_desc);
- if (desc == NULL) {
- CERROR ("Can't unpack lov_desc\n");
- GOTO (failed, rc = -EPROTO);
- }
-
- LASSERT_REPSWAB (req, 1);
- /* array of uuids byte-sex insensitive; just verify they are all
- * there and terminated */
- uuids = lustre_msg_buf (req->rq_repmsg, 1,
- desc->ld_tgt_count * sizeof (*uuids));
- if (uuids == NULL) {
- CERROR ("Can't unpack %d uuids\n", desc->ld_tgt_count);
- GOTO (failed, rc = -EPROTO);
- }
-
- for (i = 0; i < desc->ld_tgt_count; i++) {
- int uid_len = strnlen (uuids[i].uuid, sizeof (uuids[i].uuid));
-
- if (uid_len == sizeof (uuids[i].uuid)) {
- CERROR ("Unterminated uuid %d:%*s\n",
- i, (int)sizeof (uuids[i].uuid), uuids[i].uuid);
- GOTO (failed, rc = -EPROTO);
- }
- }
- RETURN(0);
-
- failed:
- ptlrpc_req_finished (req);
- RETURN (rc);
+ return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
+ LUSTRE_IMP_FULL, 0);
}
-int mdc_getattr_common (struct lustre_handle *conn,
- unsigned int ea_size, struct ptlrpc_request *req)
+/*
+ * This function now is known to always saying that it will receive 4 buffers
+ * from server. Even for cases when acl_size and md_size is zero, RPC header
+ * willcontain 4 fields and RPC itself will contain zero size fields. This is
+ * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
+ * and thus zero, it shirinks it, making zero size. The same story about
+ * md_size. And this is course of problem when client waits for smaller number
+ * of fields. This issue will be fixed later when client gets awar of RPC
+ * layouts. --umka
+ */
+static int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
+ unsigned int acl_size, int mdscapa,
+ struct ptlrpc_request *req)
{
- struct mds_body *body;
- void *eadata;
- int rc;
- int size[2] = {sizeof(*body), 0};
- int bufcount = 1;
+ struct mdt_body *body;
+ void *eadata;
+ int size[5] = { sizeof(struct ptlrpc_body),
+ sizeof(*body),
+ ea_size,
+ acl_size,
+ sizeof(struct lustre_capa) };
+ int offset, rc;
ENTRY;
- /* request message already built */
-
- if (ea_size != 0) {
- size[bufcount++] = ea_size;
+ /* Request message already built. */
+ if (ea_size)
CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
ea_size);
- }
- req->rq_replen = lustre_msg_size(bufcount, size);
+ if (acl_size)
+ CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
+
+ ptlrpc_req_set_repsize(req, 5, size);
- mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
rc = ptlrpc_queue_wait(req);
- mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
if (rc != 0)
RETURN (rc);
- body = lustre_swab_repbuf (req, 0, sizeof (*body),
- lustre_swab_mds_body);
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
if (body == NULL) {
- CERROR ("Can't unpack mds_body\n");
+ CERROR ("Can't unpack mdt_body\n");
RETURN (-EPROTO);
}
CDEBUG(D_NET, "mode: %o\n", body->mode);
- LASSERT_REPSWAB (req, 1);
+ offset = REPLY_REC_OFF + 1;
+ lustre_set_rep_swabbed(req, offset);
if (body->eadatasize != 0) {
/* reply indicates presence of eadata; check it's there... */
- eadata = lustre_msg_buf (req->rq_repmsg, 1, body->eadatasize);
+ eadata = lustre_msg_buf(req->rq_repmsg, offset++,
+ body->eadatasize);
if (eadata == NULL) {
CERROR ("Missing/short eadata\n");
RETURN (-EPROTO);
}
}
+ if (body->valid & OBD_MD_FLMODEASIZE) {
+ struct client_obd *cli = &exp->exp_obd->u.cli;
+
+ if (cli->cl_max_mds_easize < body->max_mdsize)
+ cli->cl_max_mds_easize = body->max_mdsize;
+ if (cli->cl_max_mds_cookiesize < body->max_cookiesize)
+ cli->cl_max_mds_cookiesize = body->max_cookiesize;
+ }
+
+ offset += !!body->aclsize;
+
+ if (body->valid & OBD_MD_FLMDSCAPA) {
+ struct lustre_capa *capa;
+
+ LASSERT(mdscapa);
+ capa = lustre_unpack_capa(req->rq_repmsg, offset++);
+ if (capa == NULL) {
+ CERROR("Missing/short client MDS capability\n");
+ RETURN(-EPROTO);
+ }
+ }
+
RETURN (0);
}
-int mdc_getattr(struct lustre_handle *conn, struct ll_fid *fid,
- unsigned long valid, unsigned int ea_size,
+int mdc_getattr(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, obd_valid valid, int ea_size,
struct ptlrpc_request **request)
{
struct ptlrpc_request *req;
- struct mds_body *body;
- int size = sizeof(*body);
- int rc;
+ int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
+ int acl_size = 0, rc;
ENTRY;
- /* XXX do we need to make another request here? We just did a getattr
- * to do the lookup in the first place.
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+
+ /*
+ * XXX: Do we need to make another request here? We just did a getattr
+ * to do the lookup in the first place.
*/
- req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR, 1, &size,
- NULL);
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_GETATTR, 3, size, NULL);
if (!req)
GOTO(out, rc = -ENOMEM);
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
- memcpy(&body->fid1, fid, sizeof(*fid));
- body->valid = valid;
- body->eadatasize = ea_size;
- mdc_pack_req_body(req);
+ mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, ea_size,
+ MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
- rc = mdc_getattr_common (conn, ea_size, req);
+ if (valid & OBD_MD_FLRMTPERM)
+ acl_size = sizeof(struct mdt_remote_perm);
+
+ /* Currently only root inode will call us with FLACL */
+ else if (valid & OBD_MD_FLACL)
+ acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
+
+ rc = mdc_getattr_common(exp, ea_size, acl_size,
+ !!(valid & OBD_MD_FLMDSCAPA), req);
if (rc != 0) {
ptlrpc_req_finished (req);
req = NULL;
RETURN (rc);
}
-int mdc_getattr_name(struct lustre_handle *conn, struct ll_fid *fid,
- char *filename, int namelen, unsigned long valid,
- unsigned int ea_size, struct ptlrpc_request **request)
+int mdc_getattr_name(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, const char *filename, int namelen,
+ obd_valid valid, int ea_size,
+ struct ptlrpc_request **request)
{
struct ptlrpc_request *req;
- struct mds_body *body;
- int rc, size[2] = {sizeof(*body), namelen};
+ struct mdt_body *body;
+ int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body), 0, namelen};
+ int rc;
ENTRY;
- req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_GETATTR_NAME, 2,
- size, NULL);
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_GETATTR_NAME, 4, size, NULL);
if (!req)
GOTO(out, rc = -ENOMEM);
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
- memcpy(&body->fid1, fid, sizeof(*fid));
- body->valid = valid;
- body->eadatasize = ea_size;
- mdc_pack_req_body(req);
+ mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, ea_size,
+ MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
- LASSERT (strnlen (filename, namelen) == namelen - 1);
- memcpy(lustre_msg_buf(req->rq_reqmsg, 1, namelen), filename, namelen);
+ if (filename) {
+ LASSERT(strnlen(filename, namelen) == namelen - 1);
+ memcpy(lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, namelen),
+ filename, namelen);
+ }
- rc = mdc_getattr_common (conn, ea_size, req);
+ rc = mdc_getattr_common(exp, ea_size, 0, !!(valid & OBD_MD_FLMDSCAPA),
+ req);
if (rc != 0) {
ptlrpc_req_finished (req);
req = NULL;
}
out:
*request = req;
+ RETURN(rc);
+}
+
+static int mdc_is_subdir(struct obd_export *exp, const struct lu_fid *pfid,
+ const struct lu_fid *cfid, struct ptlrpc_request **request)
+{
+ int size[2] = { sizeof(struct ptlrpc_body),
+ sizeof(struct mdt_body) };
+ struct ptlrpc_request *req;
+ struct mdt_body *body;
+ int rc;
+ ENTRY;
+
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_IS_SUBDIR, 2, size, NULL);
+ if (!req)
+ GOTO(out, rc = -ENOMEM);
+
+ mdc_is_subdir_pack(req, REQ_REC_OFF, pfid, cfid, 0);
+
+ ptlrpc_req_set_repsize(req, 2, size);
+ rc = ptlrpc_queue_wait(req);
+ if (rc != 0 && rc != -EREMOTE)
+ GOTO(out, rc);
+
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
+ if (body == NULL) {
+ CERROR ("Can't unpack mdt_body\n");
+ GOTO(out, rc = -EPROTO);
+ }
+ EXIT;
+ out:
+ *request = req;
return rc;
}
-/* This should be called with both the request and the reply still packed. */
-void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
- int repoff)
+static
+int mdc_xattr_common(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc,
+ int opcode, obd_valid valid, const char *xattr_name,
+ const char *input, int input_size, int output_size,
+ int flags, struct ptlrpc_request **request)
+{
+ struct ptlrpc_request *req;
+ int size[5] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
+ int bufcnt = 3, offset = REQ_REC_OFF + 2;
+ int rc, xattr_namelen = 0, remote_acl = 0;
+ void *tmp;
+ ENTRY;
+
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+ if (xattr_name) {
+ xattr_namelen = strlen(xattr_name) + 1;
+ size[bufcnt++] = xattr_namelen;
+ }
+ if (input_size) {
+ LASSERT(input);
+ size[bufcnt++] = input_size;
+ }
+
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ opcode, bufcnt, size, NULL);
+ if (!req)
+ GOTO(out, rc = -ENOMEM);
+
+ /* request data */
+ mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, oc, output_size, flags);
+
+
+ if (xattr_name) {
+ tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
+ memcpy(tmp, xattr_name, xattr_namelen);
+ if (!strcmp(xattr_name, XATTR_NAME_LUSTRE_ACL))
+ remote_acl = 1;
+ }
+ if (input_size) {
+ tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
+ memcpy(tmp, input, input_size);
+ }
+
+ /* reply buffers */
+ if (opcode == MDS_GETXATTR) {
+ size[REPLY_REC_OFF] = sizeof(struct mdt_body);
+ bufcnt = 2;
+ } else {
+ bufcnt = 1;
+ }
+
+ /* we do this even output_size is 0, because server is doing that */
+ size[bufcnt++] = output_size;
+ ptlrpc_req_set_repsize(req, bufcnt, size);
+
+ /* make rpc */
+ /* NB: set remote acl doesn't need hold rpc lock, because it just
+ * send command to MDS, and when it's executed on mountpoint on MDS,
+ * another mdc_xattr_common() will be invoked there. */
+ if (opcode == MDS_SETXATTR && !remote_acl)
+ mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
+
+ rc = ptlrpc_queue_wait(req);
+
+ if (opcode == MDS_SETXATTR && !remote_acl)
+ mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
+
+ if (rc != 0)
+ GOTO(err_out, rc);
+
+ if (opcode == MDS_GETXATTR) {
+ struct mdt_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
+ sizeof(*body), lustre_swab_mdt_body);
+ if (body == NULL) {
+ CERROR ("Can't unpack mdt_body\n");
+ GOTO(err_out, rc = -EPROTO);
+ }
+ }
+out:
+ *request = req;
+ RETURN (rc);
+err_out:
+ ptlrpc_req_finished(req);
+ req = NULL;
+ goto out;
+}
+
+int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, obd_valid valid, const char *xattr_name,
+ const char *input, int input_size, int output_size, int flags,
+ struct ptlrpc_request **request)
+{
+ return mdc_xattr_common(exp, fid, oc, MDS_SETXATTR, valid, xattr_name,
+ input, input_size, output_size, flags, request);
+}
+
+int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, obd_valid valid, const char *xattr_name,
+ const char *input, int input_size, int output_size, int flags,
+ struct ptlrpc_request **request)
{
- struct mds_rec_create *rec =
- lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
- struct mds_body *body =
- lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
+ return mdc_xattr_common(exp, fid, oc, MDS_GETXATTR, valid, xattr_name,
+ input, input_size, output_size, flags, request);
+}
+
+#ifdef CONFIG_FS_POSIX_ACL
+static
+int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
+ struct lustre_md *md, unsigned int offset)
+{
+ struct mdt_body *body = md->body;
+ struct posix_acl *acl;
+ void *buf;
+ int rc;
+
+ if (!body->aclsize)
+ return 0;
+
+ buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
+ if (!buf) {
+ CERROR("aclsize %u, bufcount %u, bufsize %u\n",
+ body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
+ (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
+ -1 : lustre_msg_buflen(req->rq_repmsg, offset));
+ return -EPROTO;
+ }
- LASSERT (rec != NULL);
- LASSERT (body != NULL);
+ acl = posix_acl_from_xattr(buf, body->aclsize);
+ if (IS_ERR(acl)) {
+ rc = PTR_ERR(acl);
+ CERROR("convert xattr to acl: %d\n", rc);
+ return rc;
+ }
+
+ rc = posix_acl_valid(acl);
+ if (rc) {
+ CERROR("validate acl: %d\n", rc);
+ posix_acl_release(acl);
+ return rc;
+ }
- memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
- DEBUG_REQ(D_HA, req, "storing generation %x for ino "LPD64,
- rec->cr_replayfid.generation, rec->cr_replayfid.id);
+ md->posix_acl = acl;
+ return 0;
}
+#else
+#define mdc_unpack_acl(exp, req, md, offset) 0
+#endif
-int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
- struct lustre_handle *obd_import,
+int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
+ int offset, struct obd_export *dt_exp,
+ struct obd_export *md_exp,
struct lustre_md *md)
{
int rc;
md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
LASSERT (md->body != NULL);
- LASSERT_REPSWABBED (req, offset);
+ LASSERT(lustre_rep_swabbed(req, offset));
+ offset++;
if (md->body->valid & OBD_MD_FLEASIZE) {
int lmmsize;
struct lov_mds_md *lmm;
- LASSERT(S_ISREG(md->body->mode));
+ if (!S_ISREG(md->body->mode)) {
+ CERROR("OBD_MD_FLEASIZE set, should be a regular file, "
+ "but is not\n");
+ GOTO(out, rc = -EPROTO);
+ }
if (md->body->eadatasize == 0) {
- CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
- RETURN(-EPROTO);
+ CERROR("OBD_MD_FLEASIZE set, but eadatasize 0\n");
+ GOTO(out, rc = -EPROTO);
}
lmmsize = md->body->eadatasize;
- lmm = lustre_msg_buf(req->rq_repmsg, offset + 1, lmmsize);
- LASSERT (lmm != NULL);
- LASSERT_REPSWABBED (req, offset + 1);
-
- rc = obd_unpackmd(obd_import, &md->lsm, lmm, lmmsize);
- if (rc < 0) {
- /* XXX don't know if I should do this... */
- CERROR ("Error %d unpacking eadata\n", rc);
- LBUG();
+ lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
+ if (!lmm) {
+ CERROR ("incorrect message: lmm == 0\n");
+ GOTO(out, rc = -EPROTO);
}
- LASSERT (rc >= sizeof (*md->lsm));
- }
- RETURN(0);
-}
-
+ LASSERT(lustre_rep_swabbed(req, offset));
-/* We always reserve enough space in the reply packet for a stripe MD, because
- * we don't know in advance the file type. */
-int mdc_enqueue(struct lustre_handle *conn,
- int lock_type,
- struct lookup_intent *it,
- int lock_mode,
- struct mdc_op_data *data,
- struct lustre_handle *lockh,
- char *tgt,
- int tgtlen,
- ldlm_completion_callback cb_completion,
- ldlm_blocking_callback cb_blocking,
- void *cb_data)
-{
- struct ptlrpc_request *req;
- struct obd_device *obddev = class_conn2obd(conn);
- struct ldlm_res_id res_id =
- { .name = {data->ino1, data->gen1} };
- int size[6] = {sizeof(struct ldlm_request), sizeof(struct ldlm_intent)};
- int rc, flags = LDLM_FL_HAS_INTENT;
- int repsize[4] = {sizeof(struct ldlm_reply),
- sizeof(struct mds_body),
- obddev->u.cli.cl_max_mds_easize,
- obddev->u.cli.cl_max_mds_cookiesize};
- struct ldlm_reply *dlm_rep;
- struct ldlm_intent *lit;
- struct ldlm_request *lockreq;
- void *eadata;
- unsigned long irqflags;
- int reply_buffers = 0;
- ENTRY;
+ rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
+ if (rc < 0)
+ GOTO(out, rc);
-// LDLM_DEBUG_NOLOCK("mdsintent=%s,name=%s,dir=%lu",
-// ldlm_it2str(it->it_op), it_name, it_inode->i_ino);
+ if (rc < sizeof(*md->lsm)) {
+ CERROR ("lsm size too small: rc < sizeof (*md->lsm) "
+ "(%d < %d)\n", rc, sizeof(*md->lsm));
+ GOTO(out, rc = -EPROTO);
+ }
- if (it->it_op & IT_OPEN) {
- it->it_mode |= S_IFREG;
- it->it_mode &= ~current->fs->umask;
+ offset++;
+ } else if (md->body->valid & OBD_MD_FLDIREA) {
+ int lmvsize;
+ struct lov_mds_md *lmv;
- size[2] = sizeof(struct mds_rec_create);
- size[3] = data->namelen + 1;
- req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
- size, NULL);
- if (!req)
- RETURN(-ENOMEM);
+ if(!S_ISDIR(md->body->mode)) {
+ CERROR("OBD_MD_FLDIREA set, should be a directory, but "
+ "is not\n");
+ GOTO(out, rc = -EPROTO);
+ }
- spin_lock_irqsave (&req->rq_lock, irqflags);
- req->rq_replay = 1;
- spin_unlock_irqrestore (&req->rq_lock, irqflags);
-
- /* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
- lit->opc = (__u64)it->it_op;
-
- /* pack the intended request */
- mdc_open_pack(req, 2, data, it->it_mode, 0, current->fsuid,
- current->fsgid, LTIME_S(CURRENT_TIME),
- it->it_flags, tgt, tgtlen);
- /* get ready for the reply */
- reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
- } else if (it->it_op & IT_UNLINK) {
- size[2] = sizeof(struct mds_rec_unlink);
- size[3] = data->namelen + 1;
- req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
- size, NULL);
- if (!req)
- RETURN(-ENOMEM);
+ if (md->body->eadatasize == 0) {
+ CERROR("OBD_MD_FLDIREA is set, but eadatasize 0\n");
+ RETURN(-EPROTO);
+ }
+ if (md->body->valid & OBD_MD_MEA) {
+ lmvsize = md->body->eadatasize;
+ lmv = lustre_msg_buf(req->rq_repmsg, offset, lmvsize);
+ if (!lmv) {
+ CERROR ("incorrect message: lmv == 0\n");
+ GOTO(out, rc = -EPROTO);
+ }
- /* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
- lit->opc = (__u64)it->it_op;
-
- /* pack the intended request */
- mdc_unlink_pack(req, 2, data);
- /* get ready for the reply */
- reply_buffers = 4;
- req->rq_replen = lustre_msg_size(4, repsize);
- } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
- int valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
- size[2] = sizeof(struct mds_body);
- size[3] = data->namelen + 1;
-
- req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 4,
- size, NULL);
- if (!req)
- RETURN(-ENOMEM);
+ LASSERT(lustre_rep_swabbed(req, offset));
- /* pack the intent */
- lit = lustre_msg_buf(req->rq_reqmsg, 1, sizeof (*lit));
- lit->opc = (__u64)it->it_op;
-
- /* pack the intended request */
- mdc_getattr_pack(req, valid, 2, it->it_flags, data);
- /* get ready for the reply */
- reply_buffers = 3;
- req->rq_replen = lustre_msg_size(3, repsize);
- } else if (it->it_op == IT_READDIR) {
- req = ptlrpc_prep_req(class_conn2cliimp(conn), LDLM_ENQUEUE, 1,
- size, NULL);
- if (!req)
- RETURN(-ENOMEM);
+ rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
+ lmvsize);
+ if (rc < 0)
+ GOTO(out, rc);
- /* get ready for the reply */
- reply_buffers = 1;
- req->rq_replen = lustre_msg_size(1, repsize);
- } else {
- LBUG();
- RETURN(-EINVAL);
- }
-
- mdc_get_rpc_lock(&mdc_rpc_lock, it);
- rc = ldlm_cli_enqueue(conn, req, obddev->obd_namespace, NULL, res_id,
- lock_type, NULL, 0, lock_mode, &flags,
- cb_completion, cb_blocking, cb_data, lockh);
- mdc_put_rpc_lock(&mdc_rpc_lock, it);
-
- /* Similarly, if we're going to replay this request, we don't want to
- * actually get a lock, just perform the intent. */
- if (req->rq_transno || req->rq_replay) {
- lockreq = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*lockreq));
- lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
- }
-
- /* This can go when we're sure that this can never happen */
- LASSERT(rc != -ENOENT);
- if (rc == ELDLM_LOCK_ABORTED) {
- lock_mode = 0;
- memset(lockh, 0, sizeof(*lockh));
- } else if (rc != 0) {
- CERROR("ldlm_cli_enqueue: %d\n", rc);
- LASSERT (rc < 0);
- ptlrpc_req_finished(req);
- RETURN(rc);
- } else { /* rc = 0 */
- struct ldlm_lock *lock = ldlm_handle2lock(lockh);
- LASSERT(lock);
-
- /* If the server gave us back a different lock mode, we should
- * fix up our variables. */
- if (lock->l_req_mode != lock_mode) {
- ldlm_lock_addref(lockh, lock->l_req_mode);
- ldlm_lock_decref(lockh, lock_mode);
- lock_mode = lock->l_req_mode;
+ if (rc < sizeof(*md->mea)) {
+ CERROR ("size too small: rc < sizeof(*md->mea) "
+ "(%d < %d)\n", rc, sizeof(*md->mea));
+ GOTO(out, rc = -EPROTO);
+ }
+ }
+ offset++;
+ }
+ rc = 0;
+
+ /* remote permission */
+ if (md->body->valid & OBD_MD_FLRMTPERM) {
+ md->remote_perm = lustre_msg_buf(req->rq_repmsg, offset++,
+ sizeof(struct mdt_remote_perm));
+ if (!md->remote_perm) {
+ CERROR ("incorrect message: remote_perm == 0\n");
+ GOTO(out, rc = -EPROTO);
}
+ }
- LDLM_LOCK_PUT(lock);
+ /* for ACL, it's possible that FLACL is set but aclsize is zero. only
+ * when aclsize != 0 there's an actual segment for ACL in reply
+ * buffer. */
+ else if (md->body->valid & OBD_MD_FLACL) {
+ if (md->body->aclsize) {
+ rc = mdc_unpack_acl(dt_exp, req, md, offset++);
+ if (rc)
+ GOTO(out, rc);
+#ifdef CONFIG_FS_POSIX_ACL
+ } else {
+ md->posix_acl = NULL;
+#endif
+ }
}
- dlm_rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*dlm_rep));
- LASSERT(dlm_rep != NULL); /* checked by ldlm_cli_enqueue() */
- LASSERT_REPSWABBED(req, 0); /* swabbed by ldlm_cli_enqueue() */
+ if (md->body->valid & OBD_MD_FLMDSCAPA) {
+ struct obd_capa *oc = mdc_unpack_capa(req, offset++);
- it->it_disposition = (int) dlm_rep->lock_policy_res1;
- it->it_status = (int) dlm_rep->lock_policy_res2;
- it->it_lock_mode = lock_mode;
- it->it_data = req;
+ if (IS_ERR(oc))
+ GOTO(out, rc = PTR_ERR(oc));
+ md->mds_capa = oc;
+ }
- /* We know what to expect, so we do any byte flipping required here */
- LASSERT(reply_buffers == 4 || reply_buffers == 3 || reply_buffers == 1);
- if (reply_buffers >= 3) {
- struct mds_body *body;
+ if (md->body->valid & OBD_MD_FLOSSCAPA) {
+ struct obd_capa *oc = mdc_unpack_capa(req, offset++);
- body = lustre_swab_repbuf (req, 1, sizeof (*body),
- lustre_swab_mds_body);
- if (body == NULL) {
- CERROR ("Can't swab mds_body\n");
- RETURN (-EPROTO);
- }
+ if (IS_ERR(oc))
+ GOTO(out, rc = PTR_ERR(oc));
+ md->oss_capa = oc;
+ }
- if ((body->valid & OBD_MD_FLEASIZE) != 0) {
- /* The eadata is opaque; just check that it is
- * there. Eventually, obd_unpackmd() will check
- * the contents */
- eadata = lustre_swab_repbuf(req, 2, body->eadatasize,
- NULL);
- if (eadata == NULL) {
- CERROR ("Missing/short eadata\n");
- RETURN (-EPROTO);
- }
- }
+ EXIT;
+out:
+ if (rc) {
+ if (md->oss_capa)
+ free_capa(md->oss_capa);
+ if (md->mds_capa)
+ free_capa(md->mds_capa);
+#ifdef CONFIG_FS_POSIX_ACL
+ posix_acl_release(md->posix_acl);
+#endif
+ if (md->lsm)
+ obd_free_memmd(dt_exp, &md->lsm);
}
+ return rc;
+}
- RETURN(rc);
+int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
+{
+ ENTRY;
+ RETURN(0);
}
static void mdc_replay_open(struct ptlrpc_request *req)
{
- struct obd_client_handle *och = req->rq_replay_data;
- struct lustre_handle old, *file_fh = &och->och_fh;
- struct list_head *tmp;
- struct mds_body *body;
+ struct md_open_data *mod = req->rq_cb_data;
+ struct ptlrpc_request *cur, *tmp;
+ struct obd_client_handle *och;
+ struct lustre_handle old;
+ struct mdt_body *body;
+ ENTRY;
+
+ if (mod == NULL) {
+ DEBUG_REQ(D_ERROR, req,
+ "Can't properly replay without open data.");
+ EXIT;
+ return;
+ }
+
+ body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
+ LASSERT(body != NULL);
- body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
- LASSERT (body != NULL);
+ och = mod->mod_och;
+ if (och != NULL) {
+ struct lustre_handle *file_fh;
- memcpy(&old, file_fh, sizeof(old));
- CDEBUG(D_HA, "updating handle from "LPD64" to "LPD64"\n",
- file_fh->cookie, body->handle.cookie);
- memcpy(file_fh, &body->handle, sizeof(body->handle));
+ LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
- /* A few frames up, ptlrpc_replay holds the lock, so this is safe. */
- list_for_each(tmp, &req->rq_import->imp_sending_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- if (req->rq_reqmsg->opc != MDS_CLOSE)
- continue;
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
- if (memcmp(&body->handle, &old, sizeof(old)))
- continue;
+ file_fh = &och->och_fh;
+ CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
+ file_fh->cookie, body->handle.cookie);
+ old = *file_fh;
+ *file_fh = body->handle;
+ }
- DEBUG_REQ(D_HA, req, "updating close body with new fh");
- memcpy(&body->handle, file_fh, sizeof(*file_fh));
+ list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list, rq_mod_list) {
+ int opc = lustre_msg_get_opc(cur->rq_reqmsg);
+ struct mdt_epoch *epoch = NULL;
+
+ if (opc == MDS_CLOSE || opc == MDS_DONE_WRITING) {
+ epoch = lustre_msg_buf(cur->rq_reqmsg,
+ REQ_REC_OFF, sizeof(*epoch));
+ LASSERT(epoch);
+ DEBUG_REQ(D_HA, cur, "updating %s body with new fh",
+ opc == MDS_CLOSE ? "CLOSE" : "DONE_WRITING");
+ } else if (opc == MDS_REINT) {
+ struct mdt_rec_setattr *rec;
+
+ /* Check this is REINT_SETATTR. */
+ rec = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF,
+ sizeof (*rec));
+ LASSERT(rec && rec->sa_opcode == REINT_SETATTR);
+
+ epoch = lustre_msg_buf(cur->rq_reqmsg,
+ REQ_REC_OFF + 2, sizeof(*epoch));
+ LASSERT(epoch);
+ DEBUG_REQ(D_HA, cur, "updating REINT_SETATTR body "
+ "with new fh");
+ }
+ if (epoch) {
+ if (och != NULL)
+ LASSERT(!memcmp(&old, &epoch->handle,
+ sizeof(old)));
+ epoch->handle = body->handle;
+ }
}
+ EXIT;
}
-void mdc_set_open_replay_data(struct obd_client_handle *och)
+void mdc_commit_delayed(struct ptlrpc_request *req)
{
- struct ptlrpc_request *req = och->och_req;
- struct mds_rec_create *rec =
- lustre_msg_buf(req->rq_reqmsg, 2, sizeof(*rec));
- struct mds_body *body =
- lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
+ struct md_open_data *mod = req->rq_cb_data;
+ struct ptlrpc_request *cur, *tmp;
+
+ DEBUG_REQ(D_HA, req, "req committed");
+
+ if (mod == NULL)
+ return;
+
+ req->rq_cb_data = NULL;
+ req->rq_commit_cb = NULL;
+ list_del_init(&req->rq_mod_list);
+ if (req->rq_sequence) {
+ list_for_each_entry_safe(cur, tmp, &mod->mod_replay_list,
+ rq_mod_list) {
+ LASSERT(cur != LP_POISON);
+ LASSERT(cur->rq_type != LI_POISON);
+ DEBUG_REQ(D_HA, cur, "req balanced");
+ LASSERT(cur->rq_transno != 0);
+ LASSERT(cur->rq_import == req->rq_import);
+
+ /* We no longer want to preserve this for transno-
+ * unconditional replay. */
+ spin_lock(&cur->rq_lock);
+ cur->rq_replay = 0;
+ spin_unlock(&cur->rq_lock);
+ }
+ }
- LASSERT(rec != NULL);
- /* outgoing messages always in my byte order */
- LASSERT(body != NULL);
- /* incoming message in my byte order (it's been swabbed) */
- LASSERT_REPSWABBED(req, 1);
+ if (list_empty(&mod->mod_replay_list)) {
+ if (mod->mod_och != NULL)
+ mod->mod_och->och_mod = NULL;
- memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
- req->rq_replay_cb = mdc_replay_open;
- req->rq_replay_data = och;
+ OBD_FREE_PTR(mod);
+ }
}
-int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
- struct lustre_handle *fh, struct ptlrpc_request **request)
+int mdc_set_open_replay_data(struct obd_export *exp,
+ struct obd_client_handle *och,
+ struct ptlrpc_request *open_req)
{
- struct mds_body *body;
- int rc, size = sizeof(*body);
- struct ptlrpc_request *req;
+ struct md_open_data *mod;
+ struct mdt_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
+ DLM_INTENT_REC_OFF,
+ sizeof(*rec));
+ struct mdt_body *body = lustre_msg_buf(open_req->rq_repmsg,
+ DLM_REPLY_REC_OFF,
+ sizeof(*body));
+ struct obd_import *imp = open_req->rq_import;
ENTRY;
- req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
- NULL);
- if (!req)
- GOTO(out, rc = -ENOMEM);
+ LASSERT(rec != NULL);
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
- ll_ino2fid(&body->fid1, ino, 0, type);
- memcpy(&body->handle, fh, sizeof(body->handle));
+ /* Incoming message in my byte order (it's been swabbed). */
+ LASSERT(lustre_rep_swabbed(open_req, DLM_REPLY_REC_OFF));
- req->rq_replen = lustre_msg_size(0, NULL);
+ /* Outgoing messages always in my byte order. */
+ LASSERT(body != NULL);
- mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
- rc = ptlrpc_queue_wait(req);
- mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
+ /*Only the import is replayable, we set replay_open data */
+ if (och && imp->imp_replayable) {
+ OBD_ALLOC(mod, sizeof(*mod));
+ if (mod == NULL) {
+ DEBUG_REQ(D_ERROR, open_req,
+ "Can't allocate md_open_data");
+ RETURN(0);
+ }
+ CFS_INIT_LIST_HEAD(&mod->mod_replay_list);
- EXIT;
- out:
- *request = req;
- return rc;
+ spin_lock(&open_req->rq_lock);
+ if (!open_req->rq_replay) {
+ OBD_FREE(mod, sizeof(*mod));
+ spin_unlock(&open_req->rq_lock);
+ RETURN(0);
+ }
+
+ och->och_mod = mod;
+ mod->mod_och = och;
+ open_req->rq_cb_data = mod;
+ list_add_tail(&open_req->rq_mod_list, &mod->mod_replay_list);
+ open_req->rq_commit_cb = mdc_commit_delayed;
+ spin_unlock(&open_req->rq_lock);
+ }
+
+ rec->cr_fid2 = body->fid1;
+ rec->cr_ioepoch = body->ioepoch;
+ rec->cr_old_handle.cookie = body->handle.cookie;
+ open_req->rq_replay_cb = mdc_replay_open;
+ if (!fid_is_sane(&body->fid1)) {
+ DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
+ "insane fid");
+ LBUG();
+ }
+
+ DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
+ RETURN(0);
}
-int mdc_readpage(struct lustre_handle *conn, obd_id ino, int type, __u64 offset,
- struct page *page, struct ptlrpc_request **request)
+int mdc_clear_open_replay_data(struct obd_export *exp,
+ struct obd_client_handle *och)
{
- struct obd_import *imp = class_conn2cliimp(conn);
- struct ptlrpc_request *req = NULL;
- struct ptlrpc_bulk_desc *desc = NULL;
- struct mds_body *body;
- int rc, size = sizeof(*body);
+ struct md_open_data *mod = och->och_mod;
ENTRY;
- CDEBUG(D_INODE, "inode: %ld\n", (long)ino);
+ /*
+ * Don't free the structure now (it happens in mdc_commit_delayed(),
+ * after the last request is removed from its replay list),
+ * but make sure that replay doesn't poke at the och, which is about to
+ * be freed.
+ */
+ LASSERT(mod != LP_POISON);
+ if (mod != NULL)
+ mod->mod_och = NULL;
- req = ptlrpc_prep_req(imp, MDS_READPAGE, 1, &size, NULL);
- if (!req)
+ och->och_mod = NULL;
+ RETURN(0);
+}
+
+int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
+ struct md_open_data *mod, struct ptlrpc_request **request)
+{
+ struct obd_device *obd = class_exp2obd(exp);
+ int reqsize[4] = { sizeof(struct ptlrpc_body),
+ sizeof(struct mdt_epoch),
+ sizeof(struct mdt_rec_setattr)};
+ int repsize[4] = { sizeof(struct ptlrpc_body),
+ sizeof(struct mdt_body),
+ obd->u.cli.cl_max_mds_easize,
+ obd->u.cli.cl_max_mds_cookiesize };
+ struct ptlrpc_request *req;
+ int rc;
+ ENTRY;
+
+ reqsize[REQ_REC_OFF + 2] = op_data->op_capa1 ?
+ sizeof(struct lustre_capa) : 0;
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_CLOSE, 4, reqsize, NULL);
+ if (req == NULL)
GOTO(out, rc = -ENOMEM);
+
+ /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
+ * portal whose threads are not taking any DLM locks and are therefore
+ * always progressing */
/* XXX FIXME bug 249 */
req->rq_request_portal = MDS_READPAGE_PORTAL;
- desc = ptlrpc_prep_bulk_imp (req, BULK_PUT_SINK, MDS_BULK_PORTAL);
- if (desc == NULL) {
- GOTO(out, rc = -ENOMEM);
- }
- /* NB req now owns desc and will free it when it gets freed */
-
- rc = ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
- if (rc != 0)
- GOTO(out, rc);
+ /* Ensure that this close's handle is fixed up during replay. */
+ if (likely(mod != NULL))
+ list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
+ else
+ CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
- mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE, ino, type);
+ mdc_close_pack(req, REQ_REC_OFF, op_data);
+ ptlrpc_req_set_repsize(req, 4, repsize);
+ req->rq_commit_cb = mdc_commit_delayed;
+ req->rq_replay = 1;
+ LASSERT(req->rq_cb_data == NULL);
+ req->rq_cb_data = mod;
- req->rq_replen = lustre_msg_size(1, &size);
+ mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
rc = ptlrpc_queue_wait(req);
-
- if (rc == 0) {
- LASSERT (desc->bd_page_count == 1);
- body = lustre_swab_repbuf (req, 0, sizeof (*body),
- lustre_swab_mds_body);
- if (body == NULL) {
- CERROR ("Can't unpack mds_body\n");
- GOTO (out, rc = -EPROTO);
+ mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
+
+ if (req->rq_repmsg == NULL) {
+ CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
+ req->rq_status);
+ if (rc == 0)
+ rc = req->rq_status ? req->rq_status : -EIO;
+ } else if (rc == 0 || rc == -EAGAIN) {
+ rc = lustre_msg_get_status(req->rq_repmsg);
+ if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
+ DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
+ "= %d", rc);
+ if (rc > 0)
+ rc = -rc;
+ } else if (mod == NULL) {
+ if (req->rq_import->imp_replayable)
+ CERROR("Unexpected: can't find md_open_data,"
+ "but close succeeded with replayable imp"
+ "Please tell CFS.\n");
+ }
+ if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
+ sizeof(struct mdt_body),
+ lustre_swab_mdt_body)) {
+ CERROR("Error unpacking mdt_body\n");
+ rc = -EPROTO;
}
}
EXIT;
- out:
*request = req;
+ out:
+ if (rc != 0 && rc != -EAGAIN && req && req->rq_commit_cb)
+ req->rq_commit_cb(req);
+
return rc;
}
-static int mdc_iocontrol(unsigned int cmd, struct lustre_handle *conn, int len,
- void *karg, void *uarg)
+int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
+ struct md_open_data *mod)
{
- struct obd_device *obddev = class_conn2obd(conn);
- struct obd_ioctl_data *data = karg;
- struct obd_import *imp = obddev->u.cli.cl_import;
+ struct obd_device *obd = class_exp2obd(exp);
+ struct ptlrpc_request *req;
+ int size[4] = { sizeof(struct ptlrpc_body),
+ sizeof(struct mdt_epoch),
+ sizeof(struct mdt_rec_setattr)};
+ int repsize[2] = { sizeof(struct ptlrpc_body),
+ sizeof(struct mdt_body)};
+ int rc;
ENTRY;
- switch (cmd) {
- case OBD_IOC_CLIENT_RECOVER:
- RETURN(ptlrpc_recover_import(imp, data->ioc_inlbuf1));
- case IOC_OSC_SET_ACTIVE:
- RETURN(ptlrpc_set_import_active(imp, data->ioc_offset));
- default:
- CERROR("osc_ioctl(): unrecognised ioctl %#x\n", cmd);
- RETURN(-ENOTTY);
- }
+ if (op_data->op_capa1)
+ size[REQ_REC_OFF + 2] = sizeof(struct lustre_capa);
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_DONE_WRITING, 4, size, NULL);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ mdc_close_pack(req, REQ_REC_OFF, op_data);
+
+ req->rq_replay = 1;
+ req->rq_cb_data = mod;
+ req->rq_commit_cb = mdc_commit_delayed;
+ if (likely(mod != NULL))
+ list_add_tail(&req->rq_mod_list, &mod->mod_replay_list);
+ else
+ CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
+
+ ptlrpc_req_set_repsize(req, 2, repsize);
+ mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
+ rc = ptlrpc_queue_wait(req);
+ mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
+
+ /* Close the open replay sequence if an error occured or no SOM
+ * attribute update is needed. */
+ if (rc != -EAGAIN)
+ ptlrpc_close_replay_seq(req);
+
+ if (rc && rc != -EAGAIN && req->rq_commit_cb)
+ req->rq_commit_cb(req);
+
+ ptlrpc_req_finished(req);
+ RETURN(rc);
}
-static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
- unsigned long max_age)
+#ifdef HAVE_SPLIT_SUPPORT
+int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
+ const struct page *page, int offset)
+{
+ int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
+ struct obd_import *imp = class_exp2cliimp(exp);
+ struct ptlrpc_bulk_desc *desc = NULL;
+ struct ptlrpc_request *req = NULL;
+ ENTRY;
+
+ CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
+
+ req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_WRITEPAGE, 3,
+ size, NULL);
+ if (req == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ req->rq_request_portal = MDS_READPAGE_PORTAL;
+
+ desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ /* NB req now owns desc and will free it when it gets freed. */
+ ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
+ mdc_readdir_pack(req, REQ_REC_OFF, 0, offset, fid, NULL);
+
+ ptlrpc_req_set_repsize(req, 2, size);
+ rc = ptlrpc_queue_wait(req);
+ EXIT;
+out:
+ if (req != NULL)
+ ptlrpc_req_finished(req);
+ return rc;
+}
+EXPORT_SYMBOL(mdc_sendpage);
+#endif
+
+int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, __u64 offset, struct page *page,
+ struct ptlrpc_request **request)
+{
+ int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
+ struct obd_import *imp = class_exp2cliimp(exp);
+ struct ptlrpc_bulk_desc *desc = NULL;
+ struct ptlrpc_request *req = NULL;
+ struct mdt_body *body;
+ ENTRY;
+
+ CDEBUG(D_INODE, "object: "DFID"\n", PFID(fid));
+
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+ req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 3, size,
+ NULL);
+ if (req == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ /* XXX FIXME bug 249 */
+ req->rq_request_portal = MDS_READPAGE_PORTAL;
+
+ desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ /* NB req now owns desc and will free it when it gets freed */
+ ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
+ mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid, oc);
+
+ ptlrpc_req_set_repsize(req, 2, size);
+ rc = ptlrpc_queue_wait(req);
+
+ if (rc == 0) {
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
+ if (body == NULL) {
+ CERROR("Can't unpack mdt_body\n");
+ GOTO(out, rc = -EPROTO);
+ }
+
+ if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
+ CERROR ("Unexpected # bytes transferred: %d"
+ " (%ld expected)\n",
+ req->rq_bulk->bd_nob_transferred,
+ CFS_PAGE_SIZE);
+ GOTO(out, rc = -EPROTO);
+ }
+ }
+
+ EXIT;
+ out:
+ *request = req;
+ return rc;
+}
+
+static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
+ void *karg, void *uarg)
+{
+ struct obd_device *obd = exp->exp_obd;
+ struct obd_ioctl_data *data = karg;
+ struct obd_import *imp = obd->u.cli.cl_import;
+ struct llog_ctxt *ctxt;
+ int rc;
+ ENTRY;
+
+ if (!try_module_get(THIS_MODULE)) {
+ CERROR("Can't get module. Is it alive?");
+ return -EINVAL;
+ }
+ switch (cmd) {
+ case OBD_IOC_CLIENT_RECOVER:
+ rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
+ if (rc < 0)
+ GOTO(out, rc);
+ GOTO(out, rc = 0);
+ case IOC_OSC_SET_ACTIVE:
+ rc = ptlrpc_set_import_active(imp, data->ioc_offset);
+ GOTO(out, rc);
+ case OBD_IOC_PARSE: {
+ ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
+ rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
+ GOTO(out, rc);
+ }
+#ifdef __KERNEL__
+ case OBD_IOC_LLOG_INFO:
+ case OBD_IOC_LLOG_PRINT: {
+ ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
+ rc = llog_ioctl(ctxt, cmd, data);
+
+ GOTO(out, rc);
+ }
+#endif
+ case OBD_IOC_POLL_QUOTACHECK:
+ rc = lquota_poll_check(quota_interface, exp,
+ (struct if_quotacheck *)karg);
+ GOTO(out, rc);
+ default:
+ CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
+ GOTO(out, rc = -ENOTTY);
+ }
+out:
+ module_put(THIS_MODULE);
+
+ return rc;
+}
+
+int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
+ void *key, obd_count vallen, void *val,
+ struct ptlrpc_request_set *set)
+{
+ struct obd_import *imp = class_exp2cliimp(exp);
+ int rc = -EINVAL;
+
+ if (KEY_IS(KEY_INIT_RECOV)) {
+ if (vallen != sizeof(int))
+ RETURN(-EINVAL);
+ spin_lock(&imp->imp_lock);
+ imp->imp_initial_recov = *(int *)val;
+ spin_unlock(&imp->imp_lock);
+ CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
+ exp->exp_obd->obd_name, imp->imp_initial_recov);
+ RETURN(0);
+ }
+ /* Turn off initial_recov after we try all backup servers once */
+ if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
+ if (vallen != sizeof(int))
+ RETURN(-EINVAL);
+ spin_lock(&imp->imp_lock);
+ imp->imp_initial_recov_bk = *(int *)val;
+ if (imp->imp_initial_recov_bk)
+ imp->imp_initial_recov = 1;
+ spin_unlock(&imp->imp_lock);
+ CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
+ exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
+ RETURN(0);
+ }
+ if (KEY_IS(KEY_READ_ONLY)) {
+ struct ptlrpc_request *req;
+ int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
+ char *bufs[3] = { NULL, key, val };
+
+ if (vallen != sizeof(int))
+ RETURN(-EINVAL);
+
+ spin_lock(&imp->imp_lock);
+ if (*((int *)val)) {
+ imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
+ imp->imp_connect_data.ocd_connect_flags |=
+ OBD_CONNECT_RDONLY;
+ } else {
+ imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
+ imp->imp_connect_data.ocd_connect_flags &=
+ ~OBD_CONNECT_RDONLY;
+ }
+ spin_unlock(&imp->imp_lock);
+
+ req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
+ 3, size, bufs);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ ptlrpc_req_set_repsize(req, 1, NULL);
+ if (set) {
+ rc = 0;
+ ptlrpc_set_add_req(set, req);
+ ptlrpc_check_set(set);
+ } else {
+ rc = ptlrpc_queue_wait(req);
+ ptlrpc_req_finished(req);
+ }
+
+ RETURN(rc);
+ }
+ if (KEY_IS(KEY_FLUSH_CTX)) {
+ sptlrpc_import_flush_my_ctx(imp);
+ RETURN(0);
+ }
+ if (KEY_IS(KEY_MDS_CONN)) {
+ struct obd_import *imp = class_exp2cliimp(exp);
+
+ /* mds-mds import */
+ spin_lock(&imp->imp_lock);
+ imp->imp_server_timeout = 1;
+ spin_unlock(&imp->imp_lock);
+ imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
+ CDEBUG(D_OTHER|D_WARNING, "%s: timeout / 2\n", exp->exp_obd->obd_name);
+ RETURN(0);
+ }
+
+ RETURN(rc);
+}
+
+int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
+ __u32 *vallen, void *val)
+{
+ int rc = -EINVAL;
+
+ if (KEY_IS(KEY_MAX_EASIZE)) {
+ int mdsize, *max_easize;
+
+ if (*vallen != sizeof(int))
+ RETURN(-EINVAL);
+ mdsize = *(int*)val;
+ if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
+ exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
+ max_easize = val;
+ *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
+ RETURN(0);
+ }
+ if (KEY_IS(KEY_CONN_DATA)) {
+ struct obd_import *imp = class_exp2cliimp(exp);
+ struct obd_connect_data *data = val;
+
+ if (*vallen != sizeof(*data))
+ RETURN(-EINVAL);
+
+ *data = imp->imp_connect_data;
+ RETURN(0);
+ }
+
+ RETURN(rc);
+}
+
+static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
+ __u64 max_age)
{
struct ptlrpc_request *req;
struct obd_statfs *msfs;
- int rc, size = sizeof(*msfs);
+ int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
ENTRY;
/* We could possibly pass max_age in the request (as an absolute
* during mount that would help a bit). Having relative timestamps
* is not so great if request processing is slow, while absolute
* timestamps are not ideal because they need time synchronization. */
- req = ptlrpc_prep_req(obd->u.cli.cl_import, MDS_STATFS, 0, NULL, NULL);
+ req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
+ MDS_STATFS, 1, NULL, NULL);
if (!req)
RETURN(-ENOMEM);
- req->rq_replen = lustre_msg_size(1, &size);
+ ptlrpc_req_set_repsize(req, 2, size);
- mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
rc = ptlrpc_queue_wait(req);
- mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
- if (rc)
+ if (rc) {
+ /* check connection error first */
+ if (obd->u.cli.cl_import->imp_connect_error)
+ rc = obd->u.cli.cl_import->imp_connect_error;
+
GOTO(out, rc);
+ }
- msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),lustre_swab_obd_statfs);
+ msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
+ lustre_swab_obd_statfs);
if (msfs == NULL) {
CERROR("Can't unpack obd_statfs\n");
GOTO(out, rc = -EPROTO);
}
- memcpy(osfs, msfs, sizeof (*msfs));
+ memcpy(osfs, msfs, sizeof(*msfs));
EXIT;
out:
ptlrpc_req_finished(req);
return rc;
}
-static int mdc_pin(struct lustre_handle *conn, obd_id ino, __u32 gen, int type,
+static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc,
struct obd_client_handle *handle, int flag)
{
struct ptlrpc_request *req;
- struct mds_body *body;
- int rc, size = sizeof(*body);
+ struct mdt_body *body;
+ int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
ENTRY;
- req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_PIN, 1, &size, NULL);
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_PIN, 3, size, NULL);
if (req == NULL)
RETURN(-ENOMEM);
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*body));
- ll_ino2fid(&body->fid1, ino, gen, type);
+ body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof (*body));
+ body->fid1 = *fid;
body->flags = flag;
+ mdc_pack_capa(req, REQ_REC_OFF + 1, oc);
- req->rq_replen = lustre_msg_size(1, &size);
+ ptlrpc_req_set_repsize(req, 2, size);
- mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
+ mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
rc = ptlrpc_queue_wait(req);
- mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
+ mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
if (rc) {
CERROR("pin failed: %d\n", rc);
ptlrpc_req_finished(req);
RETURN(rc);
}
- body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
if (body == NULL) {
ptlrpc_req_finished(req);
RETURN(rc);
}
memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
- handle->och_req = req; /* will be dropped by unpin */
handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
+
+ OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
+ if (handle->och_mod == NULL) {
+ DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
+ RETURN(-ENOMEM);
+ }
+
+ /* will be dropped by unpin */
+ CFS_INIT_LIST_HEAD(&handle->och_mod->mod_replay_list);
+ list_add_tail(&req->rq_mod_list, &handle->och_mod->mod_replay_list);
+
RETURN(rc);
}
-static int mdc_unpin(struct lustre_handle *conn,
+static int mdc_unpin(struct obd_export *exp,
struct obd_client_handle *handle, int flag)
{
struct ptlrpc_request *req;
- struct mds_body *body;
- int rc, size = sizeof(*body);
+ struct mdt_body *body;
+ int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
ENTRY;
if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
RETURN(0);
- req = ptlrpc_prep_req(class_conn2cliimp(conn), MDS_CLOSE, 1, &size,
- NULL);
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_CLOSE, 2, size, NULL);
if (req == NULL)
RETURN(-ENOMEM);
- body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
+ body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
body->flags = flag;
- req->rq_replen = lustre_msg_size(0, NULL);
- mdc_get_rpc_lock(&mdc_rpc_lock, NULL);
+ ptlrpc_req_set_repsize(req, 1, NULL);
+ mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
rc = ptlrpc_queue_wait(req);
- mdc_put_rpc_lock(&mdc_rpc_lock, NULL);
+ mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
if (rc != 0)
CERROR("unpin failed: %d\n", rc);
ptlrpc_req_finished(req);
- ptlrpc_req_finished(handle->och_req);
+
+ LASSERT(!list_empty(&handle->och_mod->mod_replay_list));
+ req = list_entry(handle->och_mod->mod_replay_list.next,
+ typeof(*req), rq_mod_list);
+ list_del_init(&req->rq_mod_list);
+ ptlrpc_req_finished(req);
+ LASSERT(list_empty(&handle->och_mod->mod_replay_list));
+
+ OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
RETURN(rc);
}
-static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
+int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, struct ptlrpc_request **request)
{
+ struct ptlrpc_request *req;
+ int size[3] = { sizeof(struct ptlrpc_body), sizeof(struct mdt_body) };
+ int rc;
+ ENTRY;
+
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_SYNC, 3, size, NULL);
+ if (!req)
+ RETURN(rc = -ENOMEM);
+
+ mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, oc, 0, 0);
+
+ ptlrpc_req_set_repsize(req, 2, size);
+
+ rc = ptlrpc_queue_wait(req);
+ if (rc || request == NULL)
+ ptlrpc_req_finished(req);
+ else
+ *request = req;
+
+ RETURN(rc);
+}
+
+static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
+ enum obd_import_event event)
+{
+ int rc = 0;
+
+ LASSERT(imp->imp_obd == obd);
+
+ switch (event) {
+ case IMP_EVENT_DISCON: {
+#if 0
+ /* XXX Pass event up to OBDs stack. used only for FLD now */
+ rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
+#endif
+ break;
+ }
+ case IMP_EVENT_INACTIVE: {
+ struct client_obd *cli = &obd->u.cli;
+ /*
+ * Flush current sequence to make client obtain new one
+ * from server in case of disconnect/reconnect.
+ * If range is already empty then no need to flush it.
+ */
+ if (cli->cl_seq != NULL &&
+ !range_is_exhausted(&cli->cl_seq->lcs_space)) {
+ seq_client_flush(cli->cl_seq);
+ }
+
+ rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
+ break;
+ }
+ case IMP_EVENT_INVALIDATE: {
+ struct ldlm_namespace *ns = obd->obd_namespace;
+
+ ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
+
+ break;
+ }
+ case IMP_EVENT_ACTIVE: {
+ rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
+ break;
+ }
+ case IMP_EVENT_OCD:
+ rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
+ break;
+
+ default:
+ CERROR("Unknown import event %x\n", event);
+ LBUG();
+ }
+ RETURN(rc);
+}
+
+static int mdc_fid_init(struct obd_export *exp)
+{
+ struct client_obd *cli = &exp->exp_obd->u.cli;
+ char *prefix;
+ int rc;
+ ENTRY;
+
+ OBD_ALLOC_PTR(cli->cl_seq);
+ if (cli->cl_seq == NULL)
+ RETURN(-ENOMEM);
+
+ OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
+ if (prefix == NULL)
+ GOTO(out_free_seq, rc = -ENOMEM);
+
+ snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
+ exp->exp_obd->obd_name);
+
+ /* Init client side sequence-manager */
+ rc = seq_client_init(cli->cl_seq, exp,
+ LUSTRE_SEQ_METADATA,
+ prefix, NULL);
+ OBD_FREE(prefix, MAX_OBD_NAME + 5);
+ if (rc)
+ GOTO(out_free_seq, rc);
+
+ RETURN(rc);
+out_free_seq:
+ OBD_FREE_PTR(cli->cl_seq);
+ cli->cl_seq = NULL;
+ return rc;
+}
+
+static int mdc_fid_fini(struct obd_export *exp)
+{
+ struct client_obd *cli = &exp->exp_obd->u.cli;
+ ENTRY;
+
+ if (cli->cl_seq != NULL) {
+ seq_client_fini(cli->cl_seq);
+ OBD_FREE_PTR(cli->cl_seq);
+ cli->cl_seq = NULL;
+ }
+
+ RETURN(0);
+}
+
+int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
+ struct md_op_data *op_data)
+{
+ struct client_obd *cli = &exp->exp_obd->u.cli;
+ struct lu_client_seq *seq = cli->cl_seq;
+ ENTRY;
+ RETURN(seq_client_alloc_fid(seq, fid));
+}
+
+/* XXX This method is used only to clear current fid seq
+ * once fld/mds insert failed */
+static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
+{
+ struct client_obd *cli = &exp->exp_obd->u.cli;
+
+ seq_client_flush(cli->cl_seq);
+ return 0;
+}
+
+static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
+{
+ struct client_obd *cli = &obd->u.cli;
struct lprocfs_static_vars lvars;
+ int rc;
+ ENTRY;
+
+ OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
+ if (!cli->cl_rpc_lock)
+ RETURN(-ENOMEM);
+ mdc_init_rpc_lock(cli->cl_rpc_lock);
+
+ ptlrpcd_addref();
+
+ OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
+ if (!cli->cl_setattr_lock)
+ GOTO(err_rpc_lock, rc = -ENOMEM);
+ mdc_init_rpc_lock(cli->cl_setattr_lock);
+
+ OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
+ if (!cli->cl_close_lock)
+ GOTO(err_setattr_lock, rc = -ENOMEM);
+ mdc_init_rpc_lock(cli->cl_close_lock);
+ rc = client_obd_setup(obd, cfg);
+ if (rc)
+ GOTO(err_close_lock, rc);
lprocfs_init_vars(mdc, &lvars);
- return lprocfs_obd_attach(dev, lvars.obd_vars);
+ lprocfs_obd_setup(obd, lvars.obd_vars);
+ ptlrpc_lprocfs_register_obd(obd);
+
+ rc = obd_llog_init(obd, NULL, obd, 0, NULL, NULL);
+ if (rc) {
+ mdc_cleanup(obd);
+ CERROR("failed to setup llogging subsystems\n");
+ }
+
+ RETURN(rc);
+
+err_close_lock:
+ OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
+err_setattr_lock:
+ OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
+err_rpc_lock:
+ OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
+ ptlrpcd_decref();
+ RETURN(rc);
}
-static int mdc_detach(struct obd_device *dev)
+/* Initialize the default and maximum LOV EA and cookie sizes. This allows
+ * us to make MDS RPCs with large enough reply buffers to hold the
+ * maximum-sized (= maximum striped) EA and cookie without having to
+ * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
+int mdc_init_ea_size(struct obd_export *exp, int easize,
+ int def_easize, int cookiesize)
{
- return lprocfs_obd_detach(dev);
+ struct obd_device *obd = exp->exp_obd;
+ struct client_obd *cli = &obd->u.cli;
+ ENTRY;
+
+ if (cli->cl_max_mds_easize < easize)
+ cli->cl_max_mds_easize = easize;
+
+ if (cli->cl_default_mds_easize < def_easize)
+ cli->cl_default_mds_easize = def_easize;
+
+ if (cli->cl_max_mds_cookiesize < cookiesize)
+ cli->cl_max_mds_cookiesize = cookiesize;
+
+ RETURN(0);
+}
+
+static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
+{
+ int rc = 0;
+ ENTRY;
+
+ switch (stage) {
+ case OBD_CLEANUP_EARLY:
+ case OBD_CLEANUP_EXPORTS:
+ /* If we set up but never connected, the
+ client import will not have been cleaned. */
+ if (obd->u.cli.cl_import) {
+ struct obd_import *imp;
+ imp = obd->u.cli.cl_import;
+ CERROR("client import never connected\n");
+ ptlrpc_invalidate_import(imp);
+ ptlrpc_free_rq_pool(imp->imp_rq_pool);
+ class_destroy_import(imp);
+ obd->u.cli.cl_import = NULL;
+ }
+ break;
+ case OBD_CLEANUP_SELF_EXP:
+ rc = obd_llog_finish(obd, 0);
+ if (rc != 0)
+ CERROR("failed to cleanup llogging subsystems\n");
+ case OBD_CLEANUP_OBD:
+ break;
+ }
+ RETURN(rc);
+}
+
+static int mdc_cleanup(struct obd_device *obd)
+{
+ struct client_obd *cli = &obd->u.cli;
+
+ OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
+ OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
+ OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
+
+ ptlrpc_lprocfs_unregister_obd(obd);
+ lprocfs_obd_cleanup(obd);
+ ptlrpcd_decref();
+
+ return client_obd_cleanup(obd);
+}
+
+
+static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
+ struct obd_device *tgt,
+ int count, struct llog_catid *logid,
+ struct obd_uuid *uuid)
+{
+ struct llog_ctxt *ctxt;
+ int rc;
+ ENTRY;
+
+ rc = llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
+ &llog_client_ops);
+ if (rc == 0) {
+ ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
+ ctxt->loc_imp = obd->u.cli.cl_import;
+ }
+
+ rc = llog_setup(obd, llogs, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
+ &llog_client_ops);
+ if (rc == 0) {
+ ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
+ ctxt->loc_imp = obd->u.cli.cl_import;
+ }
+
+ RETURN(rc);
+}
+
+static int mdc_llog_finish(struct obd_device *obd, int count)
+{
+ int rc;
+ ENTRY;
+
+ rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
+ if (rc) {
+ CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
+ }
+ rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
+ RETURN(rc);
+}
+
+static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
+{
+ struct lustre_cfg *lcfg = buf;
+ struct lprocfs_static_vars lvars;
+ int rc = 0;
+
+ lprocfs_init_vars(mdc, &lvars);
+
+ rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
+ return(rc);
+}
+
+/* get remote permission for current user on fid */
+int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
+ struct obd_capa *oc, struct ptlrpc_request **request)
+{
+ struct ptlrpc_request *req;
+ struct mdt_body *body;
+ struct mdt_remote_perm *perm;
+ int size[5] = { sizeof(struct ptlrpc_body), sizeof(*body) };
+ int rc;
+ ENTRY;
+
+ size[REQ_REC_OFF + 1] = oc ? sizeof(struct lustre_capa) : 0;
+
+ *request = NULL;
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_GETATTR, 3, size, NULL);
+ if (!req)
+ RETURN(-ENOMEM);
+
+ mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLRMTPERM, fid, oc, 0, 0);
+
+ size[REPLY_REC_OFF + 1] = sizeof(*perm);
+ ptlrpc_req_set_repsize(req, 5, size);
+ rc = ptlrpc_queue_wait(req);
+ if (rc) {
+ ptlrpc_req_finished(req);
+ RETURN(rc);
+ }
+
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
+ LASSERT(body);
+ LASSERT(body->valid & OBD_MD_FLRMTPERM);
+
+ perm = lustre_swab_repbuf(req, REPLY_REC_OFF + 1, sizeof(*perm),
+ lustre_swab_mdt_remote_perm);
+ LASSERT(perm);
+
+ *request = req;
+ RETURN(0);
+}
+
+static int mdc_interpret_renew_capa(struct ptlrpc_request *req, void *unused,
+ int status)
+{
+ struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
+ renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
+ struct mdt_body *body = NULL;
+ struct lustre_capa *capa;
+ ENTRY;
+
+ if (status)
+ GOTO(out, capa = ERR_PTR(status));
+
+ body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
+ lustre_swab_mdt_body);
+ if (body == NULL)
+ GOTO(out, capa = ERR_PTR(-EFAULT));
+
+ if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
+ GOTO(out, capa = ERR_PTR(-ENOENT));
+
+ capa = lustre_unpack_capa(req->rq_repmsg, REPLY_REC_OFF + 1);
+ if (!capa)
+ GOTO(out, capa = ERR_PTR(-EFAULT));
+ EXIT;
+out:
+ cb(oc, capa);
+ return 0;
+}
+
+static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
+ renew_capa_cb_t cb)
+{
+ struct ptlrpc_request *req;
+ int size[5] = { sizeof(struct ptlrpc_body),
+ sizeof(struct mdt_body),
+ sizeof(struct lustre_capa) };
+ ENTRY;
+
+ req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
+ MDS_GETATTR, 3, size, NULL);
+ if (!req)
+ RETURN(-ENOMEM);
+
+ mdc_pack_req_body(req, REQ_REC_OFF, OBD_MD_FLOSSCAPA,
+ &oc->c_capa.lc_fid, oc, 0, 0);
+
+ ptlrpc_req_set_repsize(req, 5, size);
+ req->rq_async_args.pointer_arg[0] = oc;
+ req->rq_async_args.pointer_arg[1] = cb;
+ req->rq_interpret_reply = mdc_interpret_renew_capa;
+ ptlrpcd_add_req(req);
+
+ RETURN(0);
+}
+
+static int mdc_connect(const struct lu_env *env,
+ struct lustre_handle *dlm_handle,
+ struct obd_device *obd, struct obd_uuid *cluuid,
+ struct obd_connect_data *data) {
+ struct obd_import *imp = obd->u.cli.cl_import;
+
+ /* mds-mds import features */
+ if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
+ spin_lock(&imp->imp_lock);
+ imp->imp_server_timeout = 1;
+ spin_unlock(&imp->imp_lock);
+ imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
+ CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
+ obd->obd_name);
+ }
+
+ return client_connect_import(env, dlm_handle, obd, cluuid, data);
}
struct obd_ops mdc_obd_ops = {
- o_owner: THIS_MODULE,
- o_attach: mdc_attach,
- o_detach: mdc_detach,
- o_setup: client_obd_setup,
- o_cleanup: client_obd_cleanup,
- o_connect: client_import_connect,
- o_disconnect: client_import_disconnect,
- o_iocontrol: mdc_iocontrol,
- o_statfs: mdc_statfs,
- o_pin: mdc_pin,
- o_unpin: mdc_unpin,
+ .o_owner = THIS_MODULE,
+ .o_setup = mdc_setup,
+ .o_precleanup = mdc_precleanup,
+ .o_cleanup = mdc_cleanup,
+ .o_add_conn = client_import_add_conn,
+ .o_del_conn = client_import_del_conn,
+ .o_connect = mdc_connect,
+ .o_disconnect = client_disconnect_export,
+ .o_iocontrol = mdc_iocontrol,
+ .o_set_info_async = mdc_set_info_async,
+ .o_statfs = mdc_statfs,
+ .o_pin = mdc_pin,
+ .o_unpin = mdc_unpin,
+ .o_fid_init = mdc_fid_init,
+ .o_fid_fini = mdc_fid_fini,
+ .o_fid_alloc = mdc_fid_alloc,
+ .o_fid_delete = mdc_fid_delete,
+ .o_import_event = mdc_import_event,
+ .o_llog_init = mdc_llog_init,
+ .o_llog_finish = mdc_llog_finish,
+ .o_get_info = mdc_get_info,
+ .o_process_config = mdc_process_config,
+};
+
+struct md_ops mdc_md_ops = {
+ .m_getstatus = mdc_getstatus,
+ .m_change_cbdata = mdc_change_cbdata,
+ .m_close = mdc_close,
+ .m_create = mdc_create,
+ .m_done_writing = mdc_done_writing,
+ .m_enqueue = mdc_enqueue,
+ .m_getattr = mdc_getattr,
+ .m_getattr_name = mdc_getattr_name,
+ .m_intent_lock = mdc_intent_lock,
+ .m_link = mdc_link,
+ .m_is_subdir = mdc_is_subdir,
+ .m_rename = mdc_rename,
+ .m_setattr = mdc_setattr,
+ .m_setxattr = mdc_setxattr,
+ .m_getxattr = mdc_getxattr,
+ .m_sync = mdc_sync,
+ .m_readpage = mdc_readpage,
+ .m_unlink = mdc_unlink,
+ .m_cancel_unused = mdc_cancel_unused,
+ .m_init_ea_size = mdc_init_ea_size,
+ .m_set_lock_data = mdc_set_lock_data,
+ .m_lock_match = mdc_lock_match,
+ .m_get_lustre_md = mdc_get_lustre_md,
+ .m_free_lustre_md = mdc_free_lustre_md,
+ .m_set_open_replay_data = mdc_set_open_replay_data,
+ .m_clear_open_replay_data = mdc_clear_open_replay_data,
+ .m_get_remote_perm = mdc_get_remote_perm,
+ .m_renew_capa = mdc_renew_capa
};
+extern quota_interface_t mdc_quota_interface;
+
int __init mdc_init(void)
{
+ int rc;
struct lprocfs_static_vars lvars;
- mdc_init_rpc_lock(&mdc_rpc_lock);
- mdc_init_rpc_lock(&mdc_setattr_lock);
lprocfs_init_vars(mdc, &lvars);
- return class_register_type(&mdc_obd_ops, lvars.module_vars,
- LUSTRE_MDC_NAME);
+
+ request_module("lquota");
+ quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
+ init_obd_quota_ops(quota_interface, &mdc_obd_ops);
+
+ rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
+ LUSTRE_MDC_NAME, NULL);
+ if (rc && quota_interface)
+ PORTAL_SYMBOL_PUT(mdc_quota_interface);
+
+ RETURN(rc);
}
+#ifdef __KERNEL__
static void /*__exit*/ mdc_exit(void)
{
+ if (quota_interface)
+ PORTAL_SYMBOL_PUT(mdc_quota_interface);
+
class_unregister_type(LUSTRE_MDC_NAME);
}
-#ifdef __KERNEL__
MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
MODULE_DESCRIPTION("Lustre Metadata Client");
MODULE_LICENSE("GPL");
-EXPORT_SYMBOL(mdc_req2lustre_md);
-EXPORT_SYMBOL(mdc_getstatus);
-EXPORT_SYMBOL(mdc_getlovinfo);
-EXPORT_SYMBOL(mdc_enqueue);
-EXPORT_SYMBOL(mdc_getattr);
-EXPORT_SYMBOL(mdc_getattr_name);
-EXPORT_SYMBOL(mdc_create);
-EXPORT_SYMBOL(mdc_unlink);
-EXPORT_SYMBOL(mdc_rename);
-EXPORT_SYMBOL(mdc_link);
-EXPORT_SYMBOL(mdc_readpage);
-EXPORT_SYMBOL(mdc_setattr);
-EXPORT_SYMBOL(mdc_close);
-EXPORT_SYMBOL(mdc_set_open_replay_data);
-
-EXPORT_SYMBOL(mdc_store_inode_generation);
-
module_init(mdc_init);
module_exit(mdc_exit);
#endif