Correctly check stale fid and not start epoch if ost not support SOM

b=13537
i=umka
i=vitaly

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index ab46c21..14fa8b3 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -12,6 +12,16 @@ tbd  Sun Microsystems, Inc.
        * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
         removed cwd "./" (refer to Bugzilla 14399).
 
+Severity   : normal
+Frequency  : occasional
+Bugzilla   : 13537
+Description: Correctly check stale fid, not start epoch if ost not support SOM
+Details    : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
+             overwrited with new generated fid, but mds can anwer with one of these
+             two fids and both is not stale. setattr incorectly start epoch and
+             assume will be called done_writeting, but without SOM done_writing
+             never called.
+
 Severity   : major
 Frequency  : rare, depends on device drivers and load
 Bugzilla   : 14529

diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index fe890ca..7081ba5 100644 (file)
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -890,6 +890,10 @@ dont_check_exports:
                 export->exp_lock_replay_needed = 1;
                 spin_unlock(&export->exp_lock);
                 if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO)
+                     && (data->ocd_transno == 0))
+                        CWARN("Connect with zero transno!\n");
+
+                if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_TRANSNO)
                      && data->ocd_transno < target->obd_next_recovery_transno)
                         target->obd_next_recovery_transno = data->ocd_transno;
                 target->obd_connected_clients++;

diff --git a/lustre/llite/dcache.c b/lustre/llite/dcache.c
index 42bce6f..3ef3a35 100644 (file)
--- a/lustre/llite/dcache.c
+++ b/lustre/llite/dcache.c
@@ -156,8 +156,8 @@ void ll_intent_release(struct lookup_intent *it)
         it->it_op_release = 0;
 #endif
         /* We are still holding extra reference on a request, need to free it */
-        if (it_disposition(it, DISP_ENQ_OPEN_REF)) /* open req for llfile_open*/
-                ptlrpc_req_finished(it->d.lustre.it_data);
+        if (it_disposition(it, DISP_ENQ_OPEN_REF))
+                 ptlrpc_req_finished(it->d.lustre.it_data); /* ll_file_open */
         if (it_disposition(it, DISP_ENQ_CREATE_REF)) /* create rec */
                 ptlrpc_req_finished(it->d.lustre.it_data);
         if (it_disposition(it, DISP_ENQ_COMPLETE)) /* saved req from revalidate
@@ -359,7 +359,6 @@ int ll_revalidate_it(struct dentry *de, int lookup_flags,
 
                 rc = ll_have_md_lock(de->d_parent->d_inode,
                                      MDS_INODELOCK_UPDATE);
-        
                 RETURN(rc);
         }
 
@@ -378,18 +377,11 @@ int ll_revalidate_it(struct dentry *de, int lookup_flags,
         OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_REVALIDATE_PAUSE, 5);
         ll_frob_intent(&it, &lookup_it);
         LASSERT(it);
-
         parent = de->d_parent->d_inode;
 
-        if (it->it_op & IT_CREAT) {
-                op_data = ll_prep_md_op_data(NULL, parent, NULL,
-                                             de->d_name.name, de->d_name.len,
-                                             0, LUSTRE_OPC_CREATE, NULL);
-        } else {
-                op_data = ll_prep_md_op_data(NULL, parent, de->d_inode,
-                                             de->d_name.name, de->d_name.len,
-                                             0, LUSTRE_OPC_ANY, NULL);
-        }
+        op_data = ll_prep_md_op_data(NULL, parent, de->d_inode,
+                                     de->d_name.name, de->d_name.len,
+                                     0, LUSTRE_OPC_ANY, NULL);
         if (IS_ERR(op_data))
                 RETURN(PTR_ERR(op_data));
 
@@ -398,7 +390,7 @@ int ll_revalidate_it(struct dentry *de, int lookup_flags,
                 struct ll_inode_info *lli = ll_i2info(inode);
                 struct obd_client_handle **och_p;
                 __u64 *och_usecount;
-                
+
                 /*
                  * We used to check for MDS_INODELOCK_OPEN here, but in fact
                  * just having LOOKUP lock is enough to justify inode is the
@@ -558,7 +550,7 @@ do_lookup:
                 struct mdt_body *mdt_body;
                 struct lu_fid fid = {.f_seq = 0, .f_oid = 0, .f_ver = 0};
                 mdt_body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-                
+
                 if (de->d_inode)
                         fid = *ll_inode2fid(de->d_inode);
 

diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c
index 11b4805..fc4ed66 100644 (file)
--- a/lustre/llite/llite_lib.c
+++ b/lustre/llite/llite_lib.c
@@ -1313,13 +1313,14 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
         OBD_ALLOC_PTR(op_data);
         if (op_data == NULL)
                 RETURN(-ENOMEM);
-        
+
         memcpy(&op_data->op_attr, attr, sizeof(*attr));
 
         /* Open epoch for truncate. */
-        if (ia_valid & ATTR_SIZE)
+        if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
+            (ia_valid & ATTR_SIZE))
                 op_data->op_flags = MF_EPOCH_OPEN;
-        
+
         rc = ll_md_setattr(inode, op_data, &mod);
         if (rc)
                 GOTO(out, rc);

diff --git a/lustre/lmv/lmv_intent.c b/lustre/lmv/lmv_intent.c
index 6b0c365..f87d2f5 100644 (file)
--- a/lustre/lmv/lmv_intent.c
+++ b/lustre/lmv/lmv_intent.c
@@ -243,6 +243,8 @@ repeat:
                  * For open with IT_CREATE and for IT_CREATE cases allocate new
                  * fid and setup FLD for it.
                  */
+                /* save old child fid for correctly check stale data*/
+                sop_data->op_fid3 = sop_data->op_fid2;
                 rc = lmv_fid_alloc(exp, &sop_data->op_fid2, sop_data);
                 if (rc)
                         GOTO(out_free_sop_data, rc);

diff --git a/lustre/mdc/mdc_locks.c b/lustre/mdc/mdc_locks.c
index c546ef8..343b918 100644 (file)
--- a/lustre/mdc/mdc_locks.c
+++ b/lustre/mdc/mdc_locks.c
@@ -816,8 +816,17 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
                 it_set_disposition(it, DISP_ENQ_COMPLETE);
 
                 /* Also: did we find the same inode? */
-                if (!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1))
+                /* sever can return one of two fids:
+                 * op_fid2 - new allocated fid - if file is created.
+                 * op_fid3 - existent fid - if file only open.
+                 * op_fid3 is saved in lmv_intent_open */
+                if ((!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1)) &&
+                    (!lu_fid_eq(&op_data->op_fid3, &mdt_body->fid1))) {
+                        CDEBUG(D_DENTRY, "Found stale data "DFID"("DFID")/"DFID
+                               "\n", PFID(&op_data->op_fid2),
+                               PFID(&op_data->op_fid2), PFID(&mdt_body->fid1));
                         RETURN(-ESTALE);
+                }
         }
 
         rc = it_open_error(DISP_LOOKUP_EXECD, it);

diff --git a/lustre/ptlrpc/import.c b/lustre/ptlrpc/import.c
index 93800ce..5082bc5 100644 (file)
--- a/lustre/ptlrpc/import.c
+++ b/lustre/ptlrpc/import.c
@@ -353,12 +353,17 @@ int ptlrpc_first_transno(struct obd_import *imp, __u64 *transno)
 {
         struct ptlrpc_request *req;
         struct list_head *tmp;
-        
+
         if (list_empty(&imp->imp_replay_list))
                 return 0;
         tmp = imp->imp_replay_list.next;
         req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
         *transno = req->rq_transno;
+        if (req->rq_transno == 0) {
+                DEBUG_REQ(D_ERROR, req, "zero transno in replay");
+                LBUG();
+        }
+
         return 1;
 }