#include <linux/lustre_mds.h>
#include <linux/lustre_fsfilt.h>
#include <linux/lprocfs_status.h>
+#include <linux/lustre_commit_confd.h>
+
#include "mds_internal.h"
-extern int mds_get_lovtgts(struct mds_obd *obd, int tgt_count,
- struct obd_uuid *uuidarray);
-extern int mds_get_lovdesc(struct mds_obd *obd, struct lov_desc *desc);
-int mds_finish_transno(struct mds_obd *mds, struct inode *i, void *handle,
- struct ptlrpc_request *req, int rc, int disp);
-static int mds_cleanup(struct obd_device * obddev, int force, int failover);
-
-inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req)
-{
- return &req->rq_export->exp_obd->u.mds;
-}
+static int mds_cleanup(struct obd_device *obd, int flags);
static int mds_bulk_timeout(void *data)
{
snprintf(fid_name, sizeof(fid_name), "0x%lx", ino);
+ CDEBUG(D_DENTRY, "--> mds_fid2dentry: ino %lu, gen %u, sb %p\n",
+ ino, generation, mds->mds_sb);
+
/* under ext3 this is neither supposed to return bad inodes
nor NULL inodes. */
result = ll_lookup_one_len(fid_name, mds->mds_fid_de, strlen(fid_name));
if (!inode)
RETURN(ERR_PTR(-ENOENT));
- CDEBUG(D_DENTRY, "--> mds_fid2dentry: ino %lu, gen %u, sb %p\n",
- inode->i_ino, inode->i_generation, inode->i_sb);
-
if (generation && inode->i_generation != generation) {
/* we didn't find the right inode.. */
CERROR("bad inode %lu, link: %d ct: %d or generation %u/%u\n",
mds_mfd_put(mfd);
}
-/* Call with med->med_open_lock held, please. */
-static int mds_close_mfd(struct mds_file_data *mfd, struct mds_export_data *med)
+/* Close a "file descriptor" and possibly unlink an orphan from the
+ * PENDING directory.
+ *
+ * If we are being called from mds_disconnect() because the client has
+ * disappeared, then req == NULL and we do not update last_rcvd because
+ * there is nothing that could be recovered by the client at this stage
+ * (it will not even _have_ an entry in last_rcvd anymore).
+ */
+static int mds_mfd_close(struct ptlrpc_request *req, struct obd_device *obd,
+ struct mds_file_data *mfd)
{
- struct dentry *de = NULL;
-
-#ifdef CONFIG_SMP
- LASSERT(spin_is_locked(&med->med_open_lock));
-#endif
- list_del(&mfd->mfd_list);
+ struct dentry *dparent = mfd->mfd_dentry->d_parent;
+ struct inode *child_inode = mfd->mfd_dentry->d_inode;
+ char fidname[LL_FID_NAMELEN];
+ int last_orphan, fidlen, rc = 0;
+ ENTRY;
- if (mfd->mfd_dentry->d_parent) {
- LASSERT(atomic_read(&mfd->mfd_dentry->d_parent->d_count));
- de = dget(mfd->mfd_dentry->d_parent);
+ if (dparent) {
+ LASSERT(atomic_read(&dparent->d_count) > 0);
+ dparent = dget(dparent);
}
- /* this is the actual "close" */
- l_dput(mfd->mfd_dentry);
+ fidlen = ll_fid2str(fidname, child_inode->i_ino,
+ child_inode->i_generation);
- if (de)
- l_dput(de);
+ last_orphan = mds_open_orphan_dec_test(child_inode) &&
+ mds_inode_is_orphan(child_inode);
+ /* this is the actual "close" */
+ l_dput(mfd->mfd_dentry);
mds_mfd_destroy(mfd);
- RETURN(0);
-}
-static int mds_disconnect(struct lustre_handle *conn, int failover)
-{
- struct obd_export *export = class_conn2export(conn);
- int rc;
- unsigned long flags;
- ENTRY;
+ if (dparent)
+ l_dput(dparent);
- ldlm_cancel_locks_for_export(export);
+ if (last_orphan) {
+ struct mds_obd *mds = &obd->u.mds;
+ struct inode *pending_dir = mds->mds_pending_dir->d_inode;
+ struct dentry *pending_child = NULL;
+ void *handle;
- spin_lock_irqsave(&export->exp_lock, flags);
- export->exp_failover = failover;
- spin_unlock_irqrestore(&export->exp_lock, flags);
+ CDEBUG(D_ERROR, "destroying orphan object %s\n", fidname);
- rc = class_disconnect(conn, failover);
- class_export_put(export);
+ /* Sadly, there is no easy way to save pending_child from
+ * mds_reint_unlink() into mfd, so we need to re-lookup,
+ * but normally it will still be in the dcache.
+ */
+ down(&pending_dir->i_sem);
+ pending_child = lookup_one_len(fidname, mds->mds_pending_dir,
+ fidlen);
+ if (IS_ERR(pending_child))
+ GOTO(out_lock, rc = PTR_ERR(pending_child));
+ LASSERT(pending_child->d_inode != NULL);
+
+ handle = fsfilt_start(obd, pending_dir, FSFILT_OP_UNLINK, NULL);
+ if (IS_ERR(handle))
+ GOTO(out_dput, rc = PTR_ERR(handle));
+ rc = vfs_unlink(pending_dir, pending_child);
+ if (rc)
+ CERROR("error unlinking orphan %s: rc %d\n",fidname,rc);
+
+ if (req) {
+ rc = mds_finish_transno(mds, pending_dir, handle, req,
+ rc, 0);
+ } else {
+ int err = fsfilt_commit(obd, pending_dir, handle, 0);
+ if (err) {
+ CERROR("error committing orphan unlink: %d\n",
+ err);
+ if (!rc)
+ rc = err;
+ }
+ }
+ out_dput:
+ dput(pending_child);
+ out_lock:
+ up(&pending_dir->i_sem);
+ }
RETURN(rc);
}
-static void mds_destroy_export(struct obd_export *export)
+static int mds_disconnect(struct lustre_handle *conn, int flags)
{
+ struct obd_export *export = class_conn2export(conn);
struct mds_export_data *med = &export->exp_mds_data;
- struct list_head *tmp, *n;
+ struct obd_device *obd = export->exp_obd;
+ struct obd_run_ctxt saved;
int rc;
-
ENTRY;
- LASSERT(!strcmp(export->exp_obd->obd_type->typ_name,
- LUSTRE_MDS_NAME));
- /*
- * Close any open files.
- */
+ push_ctxt(&saved, &obd->u.mds.mds_ctxt, NULL);
+ /* Close any open files (which may also cause orphan unlinking). */
spin_lock(&med->med_open_lock);
- list_for_each_safe(tmp, n, &med->med_open_head) {
+ while (!list_empty(&med->med_open_head)) {
+ struct list_head *tmp = med->med_open_head.next;
struct mds_file_data *mfd =
list_entry(tmp, struct mds_file_data, mfd_list);
#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+ /* bug 1579: fix force-closing for 2.5 */
struct dentry *dentry = mfd->mfd_dentry;
+
+ list_del(&mfd->mfd_list);
+ spin_unlock(&med->med_open_lock);
+
CERROR("force closing client file handle for %*s (%s:%lu)\n",
dentry->d_name.len, dentry->d_name.name,
kdevname(dentry->d_inode->i_sb->s_dev),
dentry->d_inode->i_ino);
+ rc = mds_mfd_close(NULL, obd, mfd);
#endif
- rc = mds_close_mfd(mfd, med);
if (rc)
CDEBUG(D_INODE, "Error closing file: %d\n", rc);
+ spin_lock(&med->med_open_lock);
}
spin_unlock(&med->med_open_lock);
+ pop_ctxt(&saved, &obd->u.mds.mds_ctxt, NULL);
+ ldlm_cancel_locks_for_export(export);
if (export->exp_outstanding_reply) {
struct ptlrpc_request *req = export->exp_outstanding_reply;
unsigned long flags;
export->exp_outstanding_reply = NULL;
}
- if (!export->exp_failover)
+ if (!(flags & OBD_OPT_FAILOVER))
mds_client_free(export);
- EXIT;
+
+ rc = class_disconnect(conn, flags);
+ class_export_put(export);
+
+ RETURN(rc);
}
/*
{
lock_kernel();
lock_super(sb);
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
if (sb->s_dirt && sb->s_op && sb->s_op->write_super)
sb->s_op->write_super(sb);
+#else
+ if (sb->s_dirt && sb->s_op) {
+ if (sb->s_op->sync_fs)
+ sb->s_op->sync_fs(sb, 1);
+ else if (sb->s_op->write_super)
+ sb->s_op->write_super(sb);
+ }
+#endif
unlock_super(sb);
unlock_kernel();
}
static int mds_getstatus(struct ptlrpc_request *req)
{
+ struct obd_device *obd = req->rq_export->exp_obd;
struct mds_obd *mds = mds_req2mds(req);
struct mds_body *body;
int rc, size = sizeof(*body);
* requests if they have any. This would be fsync_super() if it
* was exported.
*/
- mds_fsync_super(mds->mds_sb);
+ fsfilt_sync(obd, mds->mds_sb);
body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
memcpy(&body->fid1, &mds->mds_rootfid, sizeof(body->fid1));
memcpy(desc, &mds->mds_lov_desc, sizeof (*desc));
tgt_count = mds->mds_lov_desc.ld_tgt_count;
- uuid0 = lustre_msg_buf (req->rq_repmsg, 1,
- tgt_count * sizeof (*uuid0));
+ uuid0 = lustre_msg_buf(req->rq_repmsg, 1, tgt_count * sizeof (*uuid0));
if (uuid0 == NULL) {
CERROR("too many targets, enlarge client buffers\n");
req->rq_status = -ENOSPC;
req->rq_status = rc;
RETURN(0);
}
+ memcpy(&mds->mds_osc_uuid, &mds->mds_lov_desc.ld_uuid,
+ sizeof(mds->mds_osc_uuid));
RETURN(0);
}
rc = fsfilt_get_md(obd, inode, lmm, lmm_size);
if (rc < 0) {
- CERROR ("Error %d reading eadata for ino %lu\n",
- rc, inode->i_ino);
+ CERROR("Error %d reading eadata for ino %lu\n",
+ rc, inode->i_ino);
} else if (rc > 0) {
body->valid |= OBD_MD_FLEASIZE;
body->eadatasize = rc;
if (inode == NULL)
RETURN(-ENOENT);
- body = lustre_msg_buf(req->rq_repmsg, reply_off, sizeof (*body));
- LASSERT (body != NULL); /* caller prepped reply */
+ body = lustre_msg_buf(req->rq_repmsg, reply_off, sizeof(*body));
+ LASSERT(body != NULL); /* caller prepped reply */
mds_pack_inode2fid(&body->fid1, inode);
mds_pack_inode2body(body, inode);
- if (S_ISREG(inode->i_mode) &&
- (reqbody->valid & OBD_MD_FLEASIZE) != 0) {
- rc = mds_pack_md(obd, req->rq_repmsg, reply_off + 1,
- body, inode);
+ if (S_ISREG(inode->i_mode) && (reqbody->valid & OBD_MD_FLEASIZE) != 0) {
+ rc = mds_pack_md(obd, req->rq_repmsg, reply_off+1, body, inode);
+
+ /* If we have LOV EA data, the OST holds size, atime, mtime */
+ if (!(body->valid & OBD_MD_FLEASIZE))
+ body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
+ OBD_MD_FLATIME | OBD_MD_FLMTIME);
} else if (S_ISLNK(inode->i_mode) &&
(reqbody->valid & OBD_MD_LINKNAME) != 0) {
- char *symname = lustre_msg_buf(req->rq_repmsg, reply_off + 1, 0);
+ char *symname = lustre_msg_buf(req->rq_repmsg, reply_off + 1,0);
int len;
LASSERT (symname != NULL); /* caller prepped reply */
rc = 0;
}
}
+
RETURN(rc);
}
ENTRY;
body = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*body));
- LASSERT (body != NULL); /* checked by caller */
- LASSERT_REQSWABBED (req, offset); /* swabbed by caller */
+ LASSERT(body != NULL); /* checked by caller */
+ LASSERT_REQSWABBED(req, offset); /* swabbed by caller */
- if (S_ISREG(inode->i_mode) &&
- (body->valid & OBD_MD_FLEASIZE) != 0) {
+ if (S_ISREG(inode->i_mode) && (body->valid & OBD_MD_FLEASIZE)) {
int rc = fsfilt_get_md(req->rq_export->exp_obd, inode, NULL, 0);
CDEBUG(D_INODE, "got %d bytes MD data for inode %lu\n",
rc, inode->i_ino);
size[bufcount] = 0;
CERROR("MD size %d larger than maximum possible %u\n",
rc, mds->mds_max_mdsize);
- } else
+ } else {
size[bufcount] = rc;
+ }
bufcount++;
- } else if (S_ISLNK (inode->i_mode) &&
- (body->valid & OBD_MD_LINKNAME) != 0) {
+ } else if (S_ISLNK(inode->i_mode) && (body->valid & OBD_MD_LINKNAME)) {
if (inode->i_size + 1 != body->eadatasize)
- CERROR ("symlink size: %Lu, reply space: %d\n",
- inode->i_size + 1, body->eadatasize);
+ CERROR("symlink size: %Lu, reply space: %d\n",
+ inode->i_size + 1, body->eadatasize);
size[bufcount] = MIN(inode->i_size + 1, body->eadatasize);
bufcount++;
CDEBUG(D_INODE, "symlink size: %Lu, reply space: %d\n",
rc = lustre_pack_msg(bufcount, size, NULL, &req->rq_replen,
&req->rq_repmsg);
if (rc) {
- CERROR("out of memoryK\n");
- req->rq_status = rc;
- GOTO(out, rc);
+ CERROR("out of memory\n");
+ GOTO(out, req->rq_status = rc);
}
EXIT;
static void reconstruct_getattr_name(int offset, struct ptlrpc_request *req,
struct lustre_handle *client_lockh)
{
+ struct mds_export_data *med = &req->rq_export->exp_mds_data;
+ struct mds_client_data *mcd = med->med_mcd;
struct obd_device *obd = req->rq_export->exp_obd;
struct mds_obd *mds = mds_req2mds(req);
struct dentry *parent, *child;
int namelen, rc = 0;
char *name;
- if (req->rq_export->exp_outstanding_reply)
- mds_steal_ack_locks(req->rq_export, req);
+ req->rq_transno = mcd->mcd_last_transno;
+ req->rq_status = mcd->mcd_last_result;
+
+ LASSERT (req->rq_export->exp_outstanding_reply);
+
+ mds_steal_ack_locks(req->rq_export, req);
+
+ if (req->rq_status)
+ return;
body = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*body));
LASSERT (body != NULL); /* checked by caller */
uc.ouc_cap = body->capability;
uc.ouc_suppgid1 = body->suppgid;
uc.ouc_suppgid2 = -1;
+
push_ctxt(&saved, &mds->mds_ctxt, &uc);
parent = mds_fid2dentry(mds, &body->fid1, NULL);
LASSERT(!IS_ERR(parent));
}
rc = mds_getattr_internal(obd, child, req, body, offset);
- req->rq_status = rc;
+ /* XXX need to handle error here */
+ LASSERT(!rc);
l_dput(child);
l_dput(parent);
}
{
struct mds_obd *mds = mds_req2mds(req);
struct obd_device *obd = req->rq_export->exp_obd;
+ struct ldlm_reply *rep = NULL;
struct obd_run_ctxt saved;
struct mds_body *body;
struct dentry *de = NULL, *dchild = NULL;
struct ldlm_res_id child_res_id = { .name = {0} };
struct lustre_handle parent_lockh;
int namesize;
- int flags = 0, rc = 0, cleanup_phase = 0, req_was_resent;
+ int flags = 0, rc = 0, cleanup_phase = 0;
char *name;
ENTRY;
/* Swab now, before anyone looks inside the request */
- body = lustre_swab_reqbuf (req, offset, sizeof (*body),
- lustre_swab_mds_body);
+ body = lustre_swab_reqbuf(req, offset, sizeof(*body),
+ lustre_swab_mds_body);
if (body == NULL) {
- CERROR ("Can't swab mds_body\n");
- GOTO (cleanup, rc = -EFAULT);
+ CERROR("Can't swab mds_body\n");
+ GOTO(cleanup, rc = -EFAULT);
}
- LASSERT_REQSWAB (req, offset + 1);
- name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
+ LASSERT_REQSWAB(req, offset + 1);
+ name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
if (name == NULL) {
- CERROR ("Can't unpack name\n");
- GOTO (cleanup, rc = -EFAULT);
+ CERROR("Can't unpack name\n");
+ GOTO(cleanup, rc = -EFAULT);
}
namesize = req->rq_reqmsg->buflens[offset + 1];
- req_was_resent = lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
- if (child_lockh->cookie) {
- LASSERT(req_was_resent);
- reconstruct_getattr_name(offset, req, child_lockh);
- RETURN(0);
- } else if (req_was_resent) {
- DEBUG_REQ(D_HA, req, "no reply for RESENT req");
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
+ struct obd_export *exp = req->rq_export;
+ if (exp->exp_outstanding_reply &&
+ exp->exp_outstanding_reply->rq_xid == req->rq_xid) {
+ reconstruct_getattr_name(offset, req, child_lockh);
+ RETURN(0);
+ }
+ DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
+ exp->exp_outstanding_reply ?
+ exp->exp_outstanding_reply->rq_xid : (u64)0);
}
LASSERT (offset == 0 || offset == 2);
- /* if requests were at offset 2, replies go back at 1 */
- if (offset)
+ /* if requests were at offset 2, the getattr reply goes back at 1 */
+ if (offset) {
+ rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
offset = 1;
+ }
uc.ouc_fsuid = body->fsuid;
uc.ouc_fsgid = body->fsgid;
uc.ouc_suppgid2 = -1;
push_ctxt(&saved, &mds->mds_ctxt, &uc);
/* Step 1: Lookup/lock parent */
+ intent_set_disposition(rep, DISP_LOOKUP_EXECD);
de = mds_fid2locked_dentry(obd, &body->fid1, NULL, LCK_PR,
&parent_lockh);
if (IS_ERR(de))
cleanup_phase = 2; /* child dentry */
if (dchild->d_inode == NULL) {
+ intent_set_disposition(rep, DISP_LOOKUP_NEG);
GOTO(cleanup, rc = -ENOENT);
+ } else {
+ intent_set_disposition(rep, DISP_LOOKUP_POS);
}
/* Step 3: Lock child */
return rc;
}
+
+static int mds_obd_statfs(struct obd_device *obd, struct obd_statfs *osfs,
+ unsigned long max_age)
+{
+ return fsfilt_statfs(obd, obd->u.mds.mds_sb, osfs);
+}
+
static int mds_statfs(struct ptlrpc_request *req)
{
struct obd_device *obd = req->rq_export->exp_obd;
- struct obd_statfs *osfs;
- int rc, size = sizeof(*osfs);
+ int rc, size = sizeof(struct obd_statfs);
ENTRY;
rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
GOTO(out, rc);
}
- osfs = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*osfs));
- rc = fsfilt_statfs(obd, obd->u.mds.mds_sb, osfs);
+ /* We call this so that we can cache a bit - 1 jiffie worth */
+ rc = obd_statfs(obd, lustre_msg_buf(req->rq_repmsg,0,size),jiffies-HZ);
if (rc) {
- CERROR("mds: statfs failed: rc %d\n", rc);
+ CERROR("mds_obd_statfs failed: rc %d\n", rc);
GOTO(out, rc);
}
static int mds_close(struct ptlrpc_request *req)
{
struct mds_export_data *med = &req->rq_export->exp_mds_data;
+ struct obd_device *obd = req->rq_export->exp_obd;
struct mds_body *body;
struct mds_file_data *mfd;
+ struct obd_run_ctxt saved;
int rc;
ENTRY;
RETURN(-ESTALE);
}
+ rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
+ if (rc) {
+ CERROR("lustre_pack_msg: rc = %d\n", rc);
+ req->rq_status = rc;
+ }
+
spin_lock(&med->med_open_lock);
- req->rq_status = mds_close_mfd(mfd, med);
+ list_del(&mfd->mfd_list);
spin_unlock(&med->med_open_lock);
+ push_ctxt(&saved, &obd->u.mds.mds_ctxt, NULL);
+ req->rq_status = mds_mfd_close(rc ? NULL : req, obd, mfd);
+ pop_ctxt(&saved, &obd->u.mds.mds_ctxt, NULL);
+
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");
req->rq_status = -ENOMEM;
RETURN(-ENOMEM);
}
- rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
- if (rc) {
- CERROR("mds: lustre_pack_msg: rc = %d\n", rc);
- req->rq_status = rc;
- }
-
mds_mfd_put(mfd);
RETURN(0);
}
GOTO (out, rc = -EFAULT);
/* body->size is actually the offset -eeb */
- if ((body->size & (PAGE_SIZE - 1)) != 0) {
+ if ((body->size & ~PAGE_MASK) != 0) {
CERROR ("offset "LPU64"not on a page boundary\n", body->size);
GOTO (out, rc = -EFAULT);
}
break;
case MDS_REINT: {
- __u32 *opcp = lustre_msg_buf (req->rq_reqmsg, 0, sizeof (*opcp));
+ __u32 *opcp = lustre_msg_buf(req->rq_reqmsg, 0, sizeof (*opcp));
__u32 opc;
- int size[2] = {sizeof(struct mds_body), mds->mds_max_mdsize};
+ int size[3] = {sizeof(struct mds_body), mds->mds_max_mdsize,
+ mds->mds_max_cookiesize};
int bufcount;
/* NB only peek inside req now; mds_reint() will swab it */
}
opc = *opcp;
if (lustre_msg_swabbed (req->rq_reqmsg))
- __swab32s (&opc);
+ __swab32s(&opc);
DEBUG_REQ(D_INODE, req, "reint %d (%s)", opc,
- (opc < sizeof (reint_names) / sizeof (reint_names[0]) ||
- reint_names[opc] == NULL) ? reint_names[opc] : "unknown opcode");
+ (opc < sizeof(reint_names) / sizeof(reint_names[0]) ||
+ reint_names[opc] == NULL) ? reint_names[opc] :
+ "unknown opcode");
OBD_FAIL_RETURN(OBD_FAIL_MDS_REINT_NET, 0);
if (opc == REINT_UNLINK)
+ bufcount = 3;
+ else if (opc == REINT_OPEN)
bufcount = 2;
else
bufcount = 1;
rc = mds_close(req);
break;
+ case MDS_PIN:
+ DEBUG_REQ(D_INODE, req, "pin");
+ OBD_FAIL_RETURN(OBD_FAIL_MDS_PIN_NET, 0);
+ rc = mds_pin(req);
+ break;
+
case OBD_PING:
DEBUG_REQ(D_INODE, req, "ping");
rc = target_handle_ping(req);
break;
+ case OBD_LOG_CANCEL:
+ CDEBUG(D_INODE, "log cancel\n");
+ OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
+ rc = -ENOTSUPP; /* la la la */
+ break;
+
case LDLM_ENQUEUE:
DEBUG_REQ(D_INODE, req, "enqueue");
OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
struct obd_device *obd = list_entry(mds, struct obd_device,
u.mds);
req->rq_repmsg->last_xid =
- le64_to_cpu (med->med_mcd->mcd_last_xid);
+ le64_to_cpu(med->med_mcd->mcd_last_xid);
if (!obd->obd_no_transno) {
req->rq_repmsg->last_committed =
*
* Also assumes for mds_last_transno that we are not modifying it (no locking).
*/
-int mds_update_server_data(struct mds_obd *mds)
+int mds_update_server_data(struct obd_device *obd)
{
+ struct mds_obd *mds = &obd->u.mds;
struct mds_server_data *msd = mds->mds_server_data;
struct file *filp = mds->mds_rcvd_filp;
struct obd_run_ctxt saved;
msd->msd_last_transno = cpu_to_le64(mds->mds_last_transno);
msd->msd_mount_count = cpu_to_le64(mds->mds_mount_count);
- CDEBUG(D_SUPER, "MDS mount_count is %Lu, last_transno is %Lu\n",
- (unsigned long long)mds->mds_mount_count,
- (unsigned long long)mds->mds_last_transno);
- rc = lustre_fwrite(filp, (char *)msd, sizeof(*msd), &off);
+ CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
+ mds->mds_mount_count, mds->mds_last_transno);
+ rc = fsfilt_write_record(obd, filp, (char *)msd, sizeof(*msd), &off);
if (rc != sizeof(*msd)) {
CERROR("error writing MDS server data: rc = %d\n", rc);
if (rc > 0)
rc = -EIO;
GOTO(out, rc);
}
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
- rc = fsync_dev(filp->f_dentry->d_inode->i_rdev);
-#else
rc = file_fsync(filp, filp->f_dentry, 1);
-#endif
if (rc)
CERROR("error flushing MDS server data: rc = %d\n", rc);
}
/* mount the file system (secretly) */
-static int mds_setup(struct obd_device *obddev, obd_count len, void *buf)
+static int mds_setup(struct obd_device *obd, obd_count len, void *buf)
{
struct obd_ioctl_data* data = buf;
- struct mds_obd *mds = &obddev->u.mds;
+ struct mds_obd *mds = &obd->u.mds;
struct vfsmount *mnt;
int rc = 0;
unsigned long page;
if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
RETURN(rc = -EINVAL);
- obddev->obd_fsops = fsfilt_get_ops(data->ioc_inlbuf2);
- if (IS_ERR(obddev->obd_fsops))
- RETURN(rc = PTR_ERR(obddev->obd_fsops));
+ if (data->ioc_inlbuf4)
+ obd_str2uuid(&mds->mds_osc_uuid, data->ioc_inlbuf4);
+
+ obd->obd_fsops = fsfilt_get_ops(data->ioc_inlbuf2);
+ if (IS_ERR(obd->obd_fsops))
+ RETURN(rc = PTR_ERR(obd->obd_fsops));
if (data->ioc_inllen3 > 0 && data->ioc_inlbuf3) {
spin_lock_init(&mds->mds_transno_lock);
mds->mds_max_mdsize = sizeof(struct lov_mds_md);
- rc = mds_fs_setup(obddev, mnt);
+ mds->mds_max_cookiesize = sizeof(struct llog_cookie);
+ rc = mds_fs_setup(obd, mnt);
if (rc) {
CERROR("MDS filesystem method init failed: rc = %d\n", rc);
GOTO(err_put, rc);
}
- obddev->obd_namespace =
- ldlm_namespace_new("mds_server", LDLM_NAMESPACE_SERVER);
- if (obddev->obd_namespace == NULL) {
- mds_cleanup(obddev, 0, 0);
- GOTO(err_fs, rc = -ENOMEM);
+#ifdef ENABLE_ORPHANS
+ rc = llog_start_commit_thread();
+ if (rc < 0)
+ GOTO(err_fs, rc);
+#endif
+
+#ifdef ENABLE_ORPHANS
+ mds->mds_catalog = mds_get_catalog(obd);
+ if (IS_ERR(mds->mds_catalog))
+ GOTO(err_fs, rc = PTR_ERR(mds->mds_catalog));
+#endif
+
+ obd->obd_namespace = ldlm_namespace_new("mds_server",
+ LDLM_NAMESPACE_SERVER);
+ if (obd->obd_namespace == NULL) {
+ mds_cleanup(obd, 0);
+ GOTO(err_log, rc = -ENOMEM);
}
ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
- "mds_ldlm_client", &obddev->obd_ldlm_client);
+ "mds_ldlm_client", &obd->obd_ldlm_client);
mds->mds_has_lov_desc = 0;
+ obd->obd_replayable = 1;
RETURN(0);
+err_log:
+#ifdef ENABLE_ORPHANS
+ mds_put_catalog(mds->mds_catalog);
+ /* No extra cleanup needed for llog_init_commit_thread() */
err_fs:
- mds_fs_cleanup(obddev, 0);
+#endif
+ mds_fs_cleanup(obd, 0);
err_put:
unlock_kernel();
mntput(mds->mds_vfsmnt);
mds->mds_sb = 0;
lock_kernel();
err_ops:
- fsfilt_put_ops(obddev->obd_fsops);
+ fsfilt_put_ops(obd->obd_fsops);
return rc;
}
-static int mds_cleanup(struct obd_device *obddev, int force, int failover)
+static int mds_cleanup(struct obd_device *obd, int flags)
{
- struct super_block *sb;
- struct mds_obd *mds = &obddev->u.mds;
+ struct mds_obd *mds = &obd->u.mds;
ENTRY;
- sb = mds->mds_sb;
- if (!mds->mds_sb)
+ if (mds->mds_sb == NULL)
RETURN(0);
- mds_update_server_data(mds);
- mds_fs_cleanup(obddev, failover);
+#ifdef ENABLE_ORPHANS
+ mds_put_catalog(mds->mds_catalog);
+#endif
+ if (mds->mds_osc_obd)
+ obd_disconnect(&mds->mds_osc_conn, flags);
+ mds_update_server_data(obd);
+ mds_fs_cleanup(obd, flags);
unlock_kernel();
/* 2 seems normal on mds, (may_umount() also expects 2
fwiw), but we only see 1 at this point in obdfilter. */
- if (atomic_read(&obddev->u.mds.mds_vfsmnt->mnt_count) > 2){
- CERROR("%s: mount point busy, mnt_count: %d\n",
- obddev->obd_name,
- atomic_read(&obddev->u.mds.mds_vfsmnt->mnt_count));
- }
+ if (atomic_read(&obd->u.mds.mds_vfsmnt->mnt_count) > 2)
+ CERROR("%s: mount point busy, mnt_count: %d\n", obd->obd_name,
+ atomic_read(&obd->u.mds.mds_vfsmnt->mnt_count));
mntput(mds->mds_vfsmnt);
mds->mds_sb = 0;
- ldlm_namespace_free(obddev->obd_namespace);
+ ldlm_namespace_free(obd->obd_namespace);
- if (obddev->obd_recovering)
- target_cancel_recovery_timer(obddev);
+ if (obd->obd_recovering)
+ target_cancel_recovery_timer(obd);
lock_kernel();
#ifdef CONFIG_DEV_RDONLY
dev_clear_rdonly(2);
#endif
- fsfilt_put_ops(obddev->obd_fsops);
+ fsfilt_put_ops(obd->obd_fsops);
RETURN(0);
}
remote_hdl.cookie);
}
+int intent_disposition(struct ldlm_reply *rep, int flag)
+{
+ if (!rep)
+ return 0;
+ return (rep->lock_policy_res1 & flag);
+}
+
+void intent_set_disposition(struct ldlm_reply *rep, int flag)
+{
+ if (!rep)
+ return;
+ rep->lock_policy_res1 |= flag;
+}
+
static int ldlm_intent_policy(struct ldlm_namespace *ns,
struct ldlm_lock **lockp, void *req_cookie,
ldlm_mode_t mode, int flags, void *data)
{
struct ptlrpc_request *req = req_cookie;
struct ldlm_lock *lock = *lockp;
- int rc = 0;
ENTRY;
if (!req_cookie)
/* an intent needs to be considered */
struct ldlm_intent *it;
struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
- struct mds_body *mds_body;
struct ldlm_reply *rep;
- struct lustre_handle lockh = { 0 };
+ struct lustre_handle lockh;
struct ldlm_lock *new_lock;
- int rc, offset = 2, repsize[3] = {sizeof(struct ldlm_reply),
- sizeof(struct mds_body),
- mds->mds_max_mdsize};
+ int offset = 2, repsize[4] = {sizeof(struct ldlm_reply),
+ sizeof(struct mds_body),
+ mds->mds_max_mdsize,
+ mds->mds_max_cookiesize};
- it = lustre_swab_reqbuf (req, 1, sizeof (*it),
- lustre_swab_ldlm_intent);
+ it = lustre_swab_reqbuf(req, 1, sizeof (*it),
+ lustre_swab_ldlm_intent);
if (it == NULL) {
CERROR ("Intent missing\n");
- rc = req->rq_status = -EFAULT;
- RETURN (rc);
+ req->rq_status = -EFAULT;
+ RETURN(req->rq_status);
}
LDLM_DEBUG(lock, "intent policy, opc: %s",
ldlm_it2str(it->opc));
- rc = lustre_pack_msg(3, repsize, NULL, &req->rq_replen,
- &req->rq_repmsg);
- if (rc) {
- rc = req->rq_status = -ENOMEM;
- RETURN(rc);
- }
+ req->rq_status = lustre_pack_msg(it->opc == IT_UNLINK ? 4 : 3,
+ repsize, NULL, &req->rq_replen,
+ &req->rq_repmsg);
+ if (req->rq_status)
+ RETURN(req->rq_status);
rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
- rep->lock_policy_res1 = IT_INTENT_EXEC;
+ intent_set_disposition(rep, DISP_IT_EXECD);
fixup_handle_for_resent_req(req, lock, &lockh);
switch ((long)it->opc) {
case IT_OPEN:
case IT_CREAT|IT_OPEN:
- rc = mds_reint(req, offset, &lockh);
- /* We return a dentry to the client if IT_OPEN_POS is
- * set, or if we make it to the OPEN portion of the
- * programme (which implies that we created) */
- if (!(rep->lock_policy_res1 & IT_OPEN_POS ||
- rep->lock_policy_res1 & IT_OPEN_OPEN)) {
- rep->lock_policy_res2 = rc;
+ /* XXX swab here to assert that an mds_open reint
+ * packet is following */
+ rep->lock_policy_res2 = mds_reint(req, offset, &lockh);
+ /* We abort the lock if the lookup was negative and
+ * we did not make it to the OPEN portion */
+ if (intent_disposition(rep, DISP_LOOKUP_NEG) &&
+ !intent_disposition(rep, DISP_OPEN_OPEN))
RETURN(ELDLM_LOCK_ABORTED);
- }
- break;
- case IT_UNLINK:
- rc = mds_reint(req, offset, &lockh);
- /* Don't return a lock if the unlink failed, or if we're
- * not sending back an EA */
- if (rc) {
- rep->lock_policy_res2 = rc;
- RETURN(ELDLM_LOCK_ABORTED);
- }
- if (req->rq_status != 0) {
- rep->lock_policy_res2 = req->rq_status;
- RETURN(ELDLM_LOCK_ABORTED);
- }
- mds_body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*mds_body));
- if (!(mds_body->valid & OBD_MD_FLEASIZE)) {
- rep->lock_policy_res2 = rc;
- RETURN(ELDLM_LOCK_ABORTED);
- }
break;
case IT_GETATTR:
case IT_LOOKUP:
case IT_READDIR:
- rc = mds_getattr_name(offset, req, &lockh);
+ rep->lock_policy_res2 = mds_getattr_name(offset, req,
+ &lockh);
/* FIXME: we need to sit down and decide on who should
* set req->rq_status, who should return negative and
- * positive return values, and what they all mean. */
- if (rc) {
- rep->lock_policy_res2 = rc;
+ * positive return values, and what they all mean.
+ * - replay: returns 0 & req->status is old status
+ * - otherwise: returns req->status */
+ if (!intent_disposition(rep, DISP_LOOKUP_POS) ||
+ rep->lock_policy_res2)
RETURN(ELDLM_LOCK_ABORTED);
- }
if (req->rq_status != 0) {
rep->lock_policy_res2 = req->rq_status;
RETURN(ELDLM_LOCK_ABORTED);
}
/* By this point, whatever function we called above must have
- * filled in 'lockh' or returned an error. We want to give the
- * new lock to the client instead of whatever lock it was about
- * to get. */
+ * either filled in 'lockh', been an intent replay, or returned
+ * an error. We want to allow replayed RPCs to not get a lock,
+ * since we would just drop it below anyways because lock replay
+ * is done separately by the client afterwards. For regular
+ * RPCs we want to give the new lock to the client instead of
+ * whatever lock it was about to get.
+ */
new_lock = ldlm_handle2lock(&lockh);
+ if (flags & LDLM_FL_INTENT_ONLY && !new_lock)
+ RETURN(ELDLM_LOCK_ABORTED);
+
LASSERT(new_lock != NULL);
/* If we've already given this lock to a client once, then we
RETURN(ELDLM_LOCK_REPLACED);
} else {
int size = sizeof(struct ldlm_reply);
- rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen,
- &req->rq_repmsg);
- if (rc) {
+ if (lustre_pack_msg(1, &size, NULL, &req->rq_replen,
+ &req->rq_repmsg)) {
LBUG();
RETURN(-ENOMEM);
}
}
- RETURN(rc);
+ RETURN(0);
}
int mds_attach(struct obd_device *dev, obd_count len, void *data)
}
-static int mdt_cleanup(struct obd_device *obddev, int force, int failover)
+static int mdt_cleanup(struct obd_device *obddev, int flags)
{
struct mds_obd *mds = &obddev->u.mds;
ENTRY;
/* use obd ops to offer management infrastructure */
static struct obd_ops mds_obd_ops = {
- o_owner: THIS_MODULE,
- o_attach: mds_attach,
- o_detach: mds_detach,
- o_connect: mds_connect,
- o_disconnect: mds_disconnect,
- o_setup: mds_setup,
- o_cleanup: mds_cleanup,
- o_iocontrol: mds_iocontrol,
- o_destroy_export: mds_destroy_export
+ o_owner: THIS_MODULE,
+ o_attach: mds_attach,
+ o_detach: mds_detach,
+ o_connect: mds_connect,
+ o_disconnect: mds_disconnect,
+ o_setup: mds_setup,
+ o_cleanup: mds_cleanup,
+ o_statfs: mds_obd_statfs,
+ o_iocontrol: mds_iocontrol
};
static struct obd_ops mdt_obd_ops = {
return 0;
}
-static void __exit mds_exit(void)
+static void /*__exit*/ mds_exit(void)
{
ldlm_unregister_intent();
class_unregister_type(LUSTRE_MDS_NAME);