static int mds_create_objects(struct ptlrpc_request *req, int offset,
struct mds_update_record *rec,
struct mds_obd *mds, struct obd_device *obd,
- struct inode *inode, void **handle, obd_id **ids)
+ struct dentry *dchild, void **handle,
+ obd_id **ids)
{
struct obdo *oa;
struct obd_trans_info oti = { 0 };
struct mds_body *body;
struct lov_stripe_md *lsm = NULL;
struct lov_mds_md *lmm = NULL;
+ struct inode *inode = dchild->d_inode;
void *lmm_buf;
int rc, lmm_bufsize, lmm_size;
ENTRY;
RETURN(0);
}
-
if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_ALLOC_OBDO))
GOTO(out_ids, rc = -ENOMEM);
OBD_MD_FLCTIME);
if (!(rec->ur_flags & MDS_OPEN_HAS_OBJS)) {
- /* check if things like lstripe/lfs stripe are sending us the ea */
+ /* check if things like lfs setstripe are sending us the ea */
if (rec->ur_flags & MDS_OPEN_HAS_EA) {
- rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
+ rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
mds->mds_osc_exp,
0, &lsm, rec->ur_eadata);
if (rc)
GOTO(out_oa, rc);
- }
+ } else {
+ /* Per-directory striping default code removed, because
+ * it uses the same unnamed EA storage as the directory
+ * striping for CMD. -p */
+ }
LASSERT(oa->o_gr >= FILTER_GROUP_FIRST_MDS);
rc = obd_create(mds->mds_osc_exp, oa, &lsm, &oti);
if (rc) {
mfd = NULL;
}
-#warning "XXX fixme"
+ /* #warning "XXX fixme" bug 2991 */
/* Here it used to LASSERT(mfd) if exp_outstanding_reply != NULL.
* Now that exp_outstanding_reply is a list, it's just using mfd != NULL
* to detect a re-open */
GOTO(out_dput, req->rq_status = -ENOMEM);
}
put_child = 0;
+ } else {
+ body->handle.cookie = mfd->mfd_handle.h_cookie;
+ CDEBUG(D_INODE, "resend mfd %p, cookie "LPX64"\n", mfd,
+ mfd->mfd_handle.h_cookie);
}
out_dput:
/* atomically create objects if necessary */
down(&dchild->d_inode->i_sem);
mode = dchild->d_inode->i_mode;
- if ((S_ISREG(mode) || S_ISDIR(mode)) &&
- !(body->valid & OBD_MD_FLEASIZE)) {
+ if ((S_ISREG(mode) && !(body->valid & OBD_MD_FLEASIZE)) ||
+ (S_ISDIR(mode) && !(body->valid & OBD_MD_FLDIREA))) {
rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
dchild->d_inode, 0);
if (rc) {
if (rec != NULL) {
/* no EA: create objects */
rc = mds_create_objects(req, 2, rec, mds, obd,
- dchild->d_inode, handle, &ids);
+ dchild, handle, &ids);
if (rc) {
CERROR("mds_create_objects: rc = %d\n", rc);
up(&dchild->d_inode->i_sem);
RETURN(rc);
}
+ if (S_ISREG(dchild->d_inode->i_mode) &&
+ (body->valid & OBD_MD_FLEASIZE)) {
+ rc = mds_revalidate_lov_ea(obd, dchild->d_inode,
+ req->rq_repmsg, 2);
+ if (!rc)
+ rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
+ dchild->d_inode, 0);
+ if (rc) {
+ up(&dchild->d_inode->i_sem);
+ RETURN(rc);
+ }
+ }
}
- /* If the inode has EA data, then OSTs hold size, mtime */
+ /* If the inode has no EA data, then MDSs hold size, mtime */
if (S_ISREG(dchild->d_inode->i_mode) &&
!(body->valid & OBD_MD_FLEASIZE)) {
body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
{
struct obd_device *obd = req->rq_export->exp_obd;
struct mds_body *request_body, *reply_body;
- struct obd_run_ctxt saved;
+ struct lvfs_run_ctxt saved;
int rc, size = sizeof(*reply_body);
ENTRY;
RETURN(rc);
reply_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply_body));
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
rc = mds_open_by_fid(req, &request_body->fid1, reply_body,
request_body->flags, NULL, NULL);
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
RETURN(rc);
}
RETURN(rc);
}
+static int is_mount_object(struct dentry *dparent)
+{
+ struct dentry *dchild;
+
+ if (!(dparent->d_inode->i_mode & S_ISUID))
+ return 0;
+
+ dchild = lookup_one_len(".mntinfo", dparent, strlen(".mntinfo"));
+ if (IS_ERR(dchild) || dchild == NULL)
+ return 0;
+
+ dput(dchild);
+ return 1;
+}
+
int mds_open(struct mds_update_record *rec, int offset,
struct ptlrpc_request *req, struct lustre_handle *child_lockh)
{
void *handle = NULL;
struct dentry_params dp;
struct mea *mea = NULL;
- int mea_size;
+ int mea_size, update_mode;
ENTRY;
+ DEBUG_REQ(D_INODE, req, "parent "LPU64"/%u name %*s mode %o",
+ rec->ur_fid1->id, rec->ur_fid1->generation,
+ rec->ur_namelen - 1, rec->ur_name, rec->ur_mode);
+
parent_lockh[0].cookie = 0;
parent_lockh[1].cookie = 0;
* opened this file and is only replaying the RPC, so we open the
* inode by fid (at some large expense in security). */
if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
- DEBUG_REQ(D_HA, req, "open replay, disp: "LPX64"\n",
- rep->lock_policy_res1);
+ DEBUG_REQ(D_HA, req, "open replay");
+ CDEBUG(D_HA, "open fid "LPU64"/%u name %*s mode %o\n",
+ rec->ur_fid2->id, rec->ur_fid2->generation,
+ rec->ur_namelen - 1, rec->ur_name, rec->ur_mode);
LASSERT(rec->ur_fid2->id);
}
dparent = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, parent_mode,
- parent_lockh, rec->ur_name,
+ parent_lockh, &update_mode, rec->ur_name,
rec->ur_namelen - 1,
MDS_INODELOCK_UPDATE);
if (IS_ERR(dparent)) {
/* try to retrieve MEA data for this dir */
rc = mds_get_lmv_attr(obd, dparent->d_inode, &mea, &mea_size);
+ if (rc)
+ GOTO(cleanup, rc);
if (mea != NULL) {
/* dir is already splitted, check is requested filename
* should live at this MDS or at another one */
int i;
i = mea_name2idx(mea, rec->ur_name, rec->ur_namelen - 1);
- if (mea->mea_master != i) {
- CERROR("inapropriate MDS(%d) for %s. should be %d\n",
- mea->mea_master, rec->ur_name, i);
- GOTO(cleanup, rc = -ESTALE);
+ if (mea->mea_master != mea->mea_fids[i].mds) {
+ CDEBUG(D_OTHER,
+ "%s: inapropriate MDS(%d) for %lu/%u:%s."
+ " should be %d(%d)\n", obd->obd_name,
+ mea->mea_master, dparent->d_inode->i_ino,
+ dparent->d_inode->i_generation, rec->ur_name,
+ mea->mea_fids[i].mds, i);
+ GOTO(cleanup, rc = -ERESTART);
}
}
NULL, 0, NULL, child_lockh);
#ifdef S_PDIROPS
if (parent_lockh[1].cookie != 0)
- ldlm_lock_decref(parent_lockh + 1, LCK_CW);
+ ldlm_lock_decref(parent_lockh + 1, update_mode);
#endif
ldlm_lock_decref(parent_lockh, parent_mode);
if (mea)
unsigned long ino = rec->ur_fid2->id;
struct iattr iattr;
struct inode *inode;
-
- if ((rc = mds_try_to_split_dir(obd, dparent, &mea, 0))) {
- if (rc > 0) {
- /* dir got splitted */
- GOTO(cleanup, rc = -ESTALE);
- } else {
- /* error happened during spitting */
- GOTO(cleanup, rc);
- }
+ rc = mds_try_to_split_dir(obd, dparent, &mea, 0, update_mode);
+ CDEBUG(D_OTHER, "%s: splitted %lu/%u - %d\n",
+ obd->obd_name, dparent->d_inode->i_ino,
+ dparent->d_inode->i_generation, rc);
+ if (rc > 0) {
+ /* dir got splitted */
+ GOTO(cleanup, rc = -ERESTART);
+ } else if (rc < 0) {
+ /* error happened during spitting */
+ GOTO(cleanup, rc);
}
if (!(rec->ur_flags & MDS_OPEN_CREAT)) {
}
created = 1;
- LTIME_S(iattr.ia_atime) = rec->ur_time;
- LTIME_S(iattr.ia_ctime) = rec->ur_time;
- LTIME_S(iattr.ia_mtime) = rec->ur_time;
+ LTIME_S(iattr.ia_atime) = LTIME_S(rec->ur_time);
+ LTIME_S(iattr.ia_ctime) = LTIME_S(rec->ur_time);
+ LTIME_S(iattr.ia_mtime) = LTIME_S(rec->ur_time);
- iattr.ia_uid = rec->ur_fsuid;
+ iattr.ia_uid = rec->_ur_fsuid;
if (dparent->d_inode->i_mode & S_ISGID)
iattr.ia_gid = dparent->d_inode->i_gid;
else
- iattr.ia_gid = rec->ur_fsgid;
+ iattr.ia_gid = rec->_ur_fsgid;
iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
ATTR_MTIME | ATTR_CTIME;
if (rc)
CERROR("error on parent setattr: rc = %d\n", rc);
- acc_mode = 0; /* Don't check for permissions */
+ acc_mode = 0; /* Don't check for permissions */
}
LASSERT(!mds_inode_is_orphan(dchild->d_inode));
if ((rec->ur_flags & MDS_OPEN_DIRECTORY) &&
!S_ISDIR(dchild->d_inode->i_mode))
GOTO(cleanup, rc = -ENOTDIR);
-
- if (S_ISDIR(dchild->d_inode->i_mode)) {
- if (rec->ur_flags & MDS_OPEN_CREAT || rec->ur_flags & FMODE_WRITE) {
- /*we are tryying to create or write a exist dir*/
+
+ if (S_ISDIR(dchild->d_inode->i_mode)) {
+ if (rec->ur_flags & MDS_OPEN_CREAT ||
+ rec->ur_flags & FMODE_WRITE) {
+ /*we are trying to create or write a exist dir*/
GOTO(cleanup, rc = -EISDIR);
}
if (ll_permission(dchild->d_inode, acc_mode, NULL)) {
GOTO(cleanup, rc = -EACCES);
}
- }
-
- if (rc == 0) {
- struct ldlm_res_id res_id = { . name = {0} };
- ldlm_policy_data_t policy;
- int flags = 0;
- res_id.name[0] = dchild->d_inode->i_ino;
- res_id.name[1] = dchild->d_inode->i_generation;
- policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP |
- MDS_INODELOCK_UPDATE;
- rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
- res_id, LDLM_IBITS, &policy,
- LCK_PR, &flags,
- mds_blocking_ast,
- ldlm_completion_ast, NULL, NULL,
- NULL, 0, NULL, child_lockh);
- if (rc == 0)
- cleanup_phase = 3;
+ if (is_mount_object(dchild)) {
+ CERROR("Found possible GNS mount object %*s; not "
+ "opening.\n", dchild->d_name.len,
+ dchild->d_name.name);
+ GOTO(cleanup, rc = 0); // success, but don't really open
+ }
}
/* Step 5: mds_open it */
rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
req, rc, rep ? rep->lock_policy_res1 : 0);
-
switch (cleanup_phase) {
- case 3:
- if (rc)
- ldlm_lock_decref(child_lockh, LCK_PR);
case 2:
if (rc && created) {
int err = vfs_unlink(dparent->d_inode, dchild);
err);
}
} else if (created) {
+#if 0
mds_lock_new_child(obd, dchild->d_inode, NULL);
+#endif
}
l_dput(dchild);
case 1:
l_dput(dparent);
#ifdef S_PDIROPS
if (parent_lockh[1].cookie != 0)
- ldlm_lock_decref(parent_lockh + 1, LCK_CW);
+ ldlm_lock_decref(parent_lockh + 1, update_mode);
#endif
if (rc)
ldlm_lock_decref(parent_lockh, parent_mode);
else
ptlrpc_save_lock (req, parent_lockh, parent_mode);
}
- if (rc == 0)
- atomic_inc(&mds->mds_open_count);
if (mea)
OBD_FREE(mea, mea_size);
- if ((cleanup_phase != 3) && !rc)
- rc = ENOLCK;
RETURN(rc);
}
void *handle = NULL;
struct mds_body *request_body = NULL, *reply_body = NULL;
struct dentry_params dp;
- struct lov_mds_md *lmm;
+ struct iattr iattr = { 0 };
+ struct llog_create_locks *lcl = NULL;
ENTRY;
- if (req != NULL) {
+ if (req && req->rq_reqmsg != NULL)
request_body = lustre_msg_buf(req->rq_reqmsg, 0,
sizeof(*request_body));
+ if (req && req->rq_repmsg != NULL)
reply_body = lustre_msg_buf(req->rq_repmsg, 0,
sizeof(*reply_body));
- }
fidlen = ll_fid2str(fidname, inode->i_ino, inode->i_generation);
}
if (last_orphan && unlink_orphan) {
+ struct lov_mds_md *lmm = NULL;
+ int stripe_count = 0;
LASSERT(rc == 0); /* mds_put_write_access must have succeeded */
+ if (obd->obd_recovering) {
+ CDEBUG(D_HA, "not remove orphan %s until recovery"
+ " is over\n", fidname);
+ GOTO(out, rc);
+ }
+
CDEBUG(D_HA, "destroying orphan object %s\n", fidname);
/* Sadly, there is no easy way to save pending_child from
LASSERT(pending_child->d_inode != NULL);
cleanup_phase = 2; /* dput(pending_child) when finished */
- lmm = lustre_msg_buf(req->rq_repmsg, 1, 0);
- handle = fsfilt_start_log(obd, pending_dir,
- FSFILT_OP_UNLINK, NULL,
- le32_to_cpu(lmm->lmm_stripe_count));
+ if (req != NULL && req->rq_repmsg != NULL) {
+ lmm = lustre_msg_buf(req->rq_repmsg, 1, 0);
+ stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
+ }
+
+ handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK,
+ NULL, stripe_count);
if (IS_ERR(handle)) {
rc = PTR_ERR(handle);
handle = NULL;
GOTO(cleanup, rc);
}
- if (req != NULL &&
- (reply_body->valid & OBD_MD_FLEASIZE) &&
- mds_log_op_unlink(obd, pending_child->d_inode,
- lustre_msg_buf(req->rq_repmsg, 1, 0),
- req->rq_repmsg->buflens[1],
- lustre_msg_buf(req->rq_repmsg, 2, 0),
- req->rq_repmsg->buflens[2]) > 0) {
- reply_body->valid |= OBD_MD_FLCOOKIE;
- }
-
pending_child->d_fsdata = (void *) &dp;
dp.p_inum = 0;
dp.p_ptr = req;
rc = vfs_unlink(pending_dir, pending_child);
if (rc)
CERROR("error unlinking orphan %s: rc %d\n",fidname,rc);
- } else if (mfd->mfd_mode & FMODE_WRITE && rc == 0) {
+
+ if (req != NULL && req->rq_repmsg != NULL &&
+ (reply_body->valid & OBD_MD_FLEASIZE) &&
+ mds_log_op_unlink(obd, pending_child->d_inode,
+ lmm, req->rq_repmsg->buflens[1],
+ lustre_msg_buf(req->rq_repmsg, 2, 0),
+ req->rq_repmsg->buflens[2], &lcl) > 0) {
+ reply_body->valid |= OBD_MD_FLCOOKIE;
+ }
+
+ goto out; /* Don't bother updating attrs on unlinked inode */
+ }
+
+ if (request_body != NULL && mfd->mfd_mode & FMODE_WRITE && rc == 0) {
/* Update the on-disk attributes if this was the last write
* close, and all information was provided (i.e., rc == 0)
*
* XXX this should probably be abstracted with mds_reint_setattr
*/
+
#if 0
- struct iattr iattr;
+ if (request_body->valid & OBD_MD_FLMTIME &&
+ LTIME_S(request_body->mtime) > LTIME_S(inode->i_mtime)) {
+ LTIME_S(iattr.ia_mtime) = LTIME_S(request_body->mtime);
+ iattr.ia_valid |= ATTR_MTIME;
+ }
+ if (request_body->valid & OBD_MD_FLCTIME &&
+ LTIME_S(request_body->ctime) > LTIME_S(inode->i_ctime)) {
+ LTIME_S(iattr.ia_ctime) = LTIME_S(request_body->ctime);
+ iattr.ia_valid |= ATTR_CTIME;
+ }
/* XXX can't set block count with fsfilt_setattr (!) */
- iattr.ia_valid = ATTR_CTIME | ATTR_ATIME |
- ATTR_MTIME | ATTR_SIZE;
- iattr.ia_atime = request_body->atime;
- iattr.ia_ctime = request_body->ctime;
- iattr.ia_mtime = request_body->mtime;
- iattr.ia_size = request_body->size;
- /* iattr.ia_blocks = request_body->blocks */
+ if (request_body->valid & OBD_MD_FLSIZE) {
+ iattr.ia_valid |= ATTR_SIZE;
+ iattr.ia_size = request_body->size;
+ }
+ /* if (request_body->valid & OBD_MD_FLBLOCKS) {
+ iattr.ia_valid |= ATTR_BLOCKS;
+ iattr.ia_blocks = request_body->blocks
+ } */
+
+#endif
+ }
+
+ if (request_body != NULL && request_body->valid & OBD_MD_FLATIME) {
+ /* Only start a transaction to write out only the atime if
+ * it is more out-of-date than the specified limit. If we
+ * are already going to write out the atime then do it anyway.
+ * */
+ if ((LTIME_S(request_body->atime) >
+ LTIME_S(inode->i_atime) + MAX_ATIME_DIFF) ||
+ (iattr.ia_valid != 0 &&
+ LTIME_S(request_body->atime) > LTIME_S(inode->i_atime))) {
+ LTIME_S(iattr.ia_atime) = LTIME_S(request_body->atime);
+ iattr.ia_valid |= ATTR_ATIME;
+ }
+ }
+ if (iattr.ia_valid != 0) {
handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
if (IS_ERR(handle))
GOTO(cleanup, rc = PTR_ERR(handle));
rc = fsfilt_setattr(obd, mfd->mfd_dentry, handle, &iattr, 0);
if (rc)
CERROR("error in setattr(%s): rc %d\n", fidname, rc);
-#endif
}
+out:
/* If other clients have this file open for write, rc will be > 0 */
if (rc > 0)
rc = 0;
mds_mfd_destroy(mfd);
cleanup:
- atomic_dec(&mds->mds_open_count);
- if (req) {
+ if (req != NULL && reply_body != NULL) {
rc = mds_finish_transno(mds, pending_dir, handle, req, rc, 0);
} else if (handle) {
- int err = fsfilt_commit(obd, pending_dir, handle, 0);
+ int err = fsfilt_commit(obd, mds->mds_sb, pending_dir, handle, 0);
if (err) {
CERROR("error committing close: %d\n", err);
if (!rc)
switch (cleanup_phase) {
case 2:
+ if (lcl != NULL)
+ ptlrpc_save_llog_lock(req, lcl);
dput(pending_child);
case 1:
up(&pending_dir->i_sem);
struct obd_device *obd = req->rq_export->exp_obd;
struct mds_body *body;
struct mds_file_data *mfd;
- struct obd_run_ctxt saved;
+ struct lvfs_run_ctxt saved;
struct inode *inode;
int rc, repsize[3] = {sizeof(struct mds_body),
obd->u.mds.mds_max_mdsize,
MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
}
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
+ DEBUG_REQ(D_HA, req, "close replay\n");
+ memcpy(lustre_msg_buf(req->rq_repmsg, 2, 0),
+ lustre_msg_buf(req->rq_reqmsg, 1, 0),
+ req->rq_repmsg->buflens[2]);
+ }
+
body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
if (body == NULL) {
CERROR("Can't unpack body\n");
list_del(&mfd->mfd_list);
spin_unlock(&med->med_open_lock);
- push_ctxt(&saved, &obd->obd_ctxt, NULL);
- req->rq_status = mds_mfd_close(rc ? NULL : req, obd, mfd, 1);
- pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+ req->rq_status = mds_mfd_close(req, obd, mfd, 1);
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");