Don't copy lvb into possibly NULL reply on ldlm_handle_enqueue() error

client won't use it anyways.  Regression test for same.
Fix secondary bug hit during regression tests in mds_mfd_close() cleanup.
b=2983

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index 5a2918b..8c489e1 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -1,3 +1,8 @@
+tbd  Cluster File Systems, Inc. <info@clusterfs.com>
+       * version 1.2.x
+       * Bug fixes
+       - don't copy lvb into (possibly NULL) reply on error (2983)
+
 2004-03-22  Cluster File Systems, Inc. <info@clusterfs.com>
        * version 1.2.1
        * bug fixes
@@ -21,7 +26,7 @@
        - don't oops on a deleted current working directory (2399)
        - handle hard links to targets without a parent properly (2517)
        - don't dereference NULL lock when racing during eviction (2867)
-       - don't grow lock extents when there are lots of conflicting locks (2919)
+       - don't grow lock extents when lots of conflicting locks (2919)
 
 2004-03-04  Cluster File Systems, Inc. <info@clusterfs.com>
        * version 1.2.0

diff --git a/lustre/include/linux/obd_support.h b/lustre/include/linux/obd_support.h
index c39cb6f..2daa927 100644 (file)
--- a/lustre/include/linux/obd_support.h
+++ b/lustre/include/linux/obd_support.h
@@ -113,6 +113,8 @@ extern wait_queue_head_t obd_race_waitq;
 #define OBD_FAIL_LDLM_BL_CALLBACK        0x305
 #define OBD_FAIL_LDLM_CP_CALLBACK        0x306
 #define OBD_FAIL_LDLM_GL_CALLBACK        0x307
+#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
+#define OBD_FAIL_LDLM_ENQUEUE_INTENT_ERR 0x309
 
 #define OBD_FAIL_OSC                     0x400
 #define OBD_FAIL_OSC_BRW_READ_BULK       0x401

diff --git a/lustre/ldlm/ldlm_lockd.c b/lustre/ldlm/ldlm_lockd.c
index c88139f..b456ff6 100644 (file)
--- a/lustre/ldlm/ldlm_lockd.c
+++ b/lustre/ldlm/ldlm_lockd.c
@@ -617,6 +617,9 @@ int ldlm_handle_enqueue(struct ptlrpc_request *req,
                         buffers = 2;
                 }
 
+                if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR))
+                        GOTO(out, rc = -ENOMEM);
+
                 rc = lustre_pack_reply(req, buffers, size, NULL);
                 if (rc)
                         GOTO(out, rc);
@@ -656,6 +659,7 @@ int ldlm_handle_enqueue(struct ptlrpc_request *req,
                 err = lustre_pack_reply(req, 0, NULL, NULL);
                 if (rc == 0)
                         rc = err;
+                req->rq_status = rc;
         }
 
         /* The LOCK_CHANGED code in ldlm_lock_enqueue depends on this
@@ -664,9 +668,10 @@ int ldlm_handle_enqueue(struct ptlrpc_request *req,
                 LDLM_DEBUG(lock, "server-side enqueue handler, sending reply"
                            "(err=%d, rc=%d)", err, rc);
 
-                if (lock->l_resource->lr_lvb_len > 0) {
+                if (lock->l_resource->lr_lvb_len > 0 && rc == 0) {
                         void *lvb = lustre_msg_buf(req->rq_repmsg, 1,
                                                   lock->l_resource->lr_lvb_len);
+                        LASSERT(lvb != NULL);
                         memcpy(lvb, lock->l_resource->lr_lvb_data,
                                lock->l_resource->lr_lvb_len);
                 }

diff --git a/lustre/llite/file.c b/lustre/llite/file.c
index b40f122..bc3f7f7 100644 (file)
--- a/lustre/llite/file.c
+++ b/lustre/llite/file.c
@@ -951,7 +951,7 @@ static int ll_lov_setea(struct inode *inode, struct file *file,
 {
         int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
         struct lov_user_md  *lump;
-        int lum_size = sizeof(struct lov_user_md) + 
+        int lum_size = sizeof(struct lov_user_md) +
                        sizeof(struct lov_user_ost_data);
         int rc;
         ENTRY;
@@ -963,8 +963,7 @@ static int ll_lov_setea(struct inode *inode, struct file *file,
         if (lump == NULL) {
                 RETURN(-ENOMEM);
         }
-        rc = copy_from_user(lump, (struct lov_user_md  *)arg, 
-                            lum_size);
+        rc = copy_from_user(lump, (struct lov_user_md  *)arg, lum_size);
         if (rc) {
                 OBD_FREE(lump, lum_size);
                 RETURN(-EFAULT);
@@ -1041,7 +1040,7 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
         case LL_IOC_LOV_SETSTRIPE:
                 RETURN(ll_lov_setstripe(inode, file, arg));
         case LL_IOC_LOV_SETEA:
-                RETURN( ll_lov_setea(inode, file, arg) ); 
+                RETURN(ll_lov_setea(inode, file, arg));
         case LL_IOC_LOV_GETSTRIPE:
                 RETURN(ll_lov_getstripe(inode, arg));
         case LL_IOC_RECREATE_OBJ:

diff --git a/lustre/mdc/mdc_reint.c b/lustre/mdc/mdc_reint.c
index d092270..a845347 100644 (file)
--- a/lustre/mdc/mdc_reint.c
+++ b/lustre/mdc/mdc_reint.c
@@ -41,7 +41,6 @@ static int mdc_reint(struct ptlrpc_request *request,
                      struct mdc_rpc_lock *rpc_lock, int level)
 {
         int rc;
-        
 
         request->rq_send_state = level;
 

diff --git a/lustre/mds/mds_open.c b/lustre/mds/mds_open.c
index 8b8bae0..c16b374 100644 (file)
--- a/lustre/mds/mds_open.c
+++ b/lustre/mds/mds_open.c
@@ -360,7 +360,6 @@ static int mds_create_objects(struct ptlrpc_request *req, int offset,
                 RETURN(0);
         }
 
-        
         if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_ALLOC_OBDO))
                 GOTO(out_ids, rc = -ENOMEM);
 
@@ -380,9 +379,9 @@ static int mds_create_objects(struct ptlrpc_request *req, int offset,
                         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)
@@ -952,9 +951,10 @@ int mds_open(struct mds_update_record *rec, int offset,
         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) {
+
+        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*/
                         GOTO(cleanup, rc = -EISDIR);
                 }
@@ -1021,7 +1021,6 @@ int mds_mfd_close(struct ptlrpc_request *req, struct obd_device *obd,
         void *handle = NULL;
         struct mds_body *request_body = NULL, *reply_body = NULL;
         struct dentry_params dp;
-        struct lov_mds_md *lmm;
         ENTRY;
 
         if (req != NULL) {
@@ -1045,6 +1044,7 @@ int mds_mfd_close(struct ptlrpc_request *req, struct obd_device *obd,
         }
 
         if (last_orphan && unlink_orphan) {
+                int stripe_count = 0;
                 LASSERT(rc == 0); /* mds_put_write_access must have succeeded */
 
                 CDEBUG(D_HA, "destroying orphan object %s\n", fidname);
@@ -1062,18 +1062,21 @@ int mds_mfd_close(struct ptlrpc_request *req, struct obd_device *obd,
                 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) {
+                        struct lov_mds_md *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) &&
+                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],

diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 4b2c37f..57ab438 100755 (executable)
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -2,6 +2,9 @@
 
 set -e
 
+# 17 = bug 2732   2986
+ALWAYS_EXCEPT="17 19b"
+
 LUSTRE=${LUSTRE:-`dirname $0`/..}
 UPCALL=${UPCALL:-$PWD/recovery-small-upcall.sh}
 . $LUSTRE/tests/test-framework.sh
@@ -220,8 +223,35 @@ test_19b() {
     drop_ldlm_cancel multiop $f Ow  || echo "client evicted, as expected"
 
     do_facet client munlink $f  || return 4
-
 }
 run_test 19b "test expired_lock_main on ost (2867)"
 
+test_20a() {   # bug 2983 - ldlm_handle_enqueue cleanup
+       mkdir -p $DIR/$tdir
+       multiop $DIR/$tdir/${tfile} O_wc &
+       MULTI_PID=$!
+       usleep 500
+       cancel_lru_locks OSC
+#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
+       do_facet ost sysctl -w lustre.fail_loc=0x80000308
+       set -vx
+       kill -USR1 $MULTI_PID
+       wait $MULTI_PID
+       rc=$?
+       [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
+       set +vx
+}
+run_test 20a "ldlm_handle_enqueue error (should return error)" 
+
+test_20b() {   # bug 2986 - ldlm_handle_enqueue error during open
+       mkdir $DIR/$tdir
+       touch $DIR/$tdir/${tfile}
+       cancel_lru_locks OSC
+#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
+       do_facet ost sysctl -w lustre.fail_loc=0x80000308
+       dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
+               error "didn't fail open enqueue" || true
+}
+run_test 20b "ldlm_handle_enqueue error (should return error)"
+
 $CLEANUP