dht: Rebalance causing IO Error - File descriptor in bad state

Problem : When a file is migrated, dht attempts to re-open all open
          fds on the new cached subvol. Earlier, if dht had not opened the fd,
          the client xlator would be unable to find the remote fd and would
          fall back to using an anon fd for the fop. That behavior changed with
          https://review.gluster.org/#/c/glusterfs/+/15804, causing fops to fail
          with EBADFD if the fd was not available on the cached subvol.
          The client xlator returns EBADFD if the remote fd is not found but
          dht only checks for EBADF before re-opening fds on the new cached subvol.

Solution: Handle EBADFD at dht code path to avoid the issue

Change-Id: I43c51995cdd48d05b12e4b2889c8dbe2bb2a72d8
Fixes: bz#1758579

5 files changed, 116 insertions, 17 deletions

diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index b295ce8bd03..eda01e79324 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -51,6 +51,17 @@ dht_set_dir_xattr_req(xlator_t *this, loc_t *loc, dict_t *xattr_req);
 int
 dht_do_fresh_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc);
 
+/* Check the xdata to make sure EBADF has been set by client xlator */
+int32_t
+dht_check_remote_fd_failed_error(dht_local_t *local, int op_ret, int op_errno)
+{
+    if (op_ret == -1 && (op_errno == EBADF || op_errno == EBADFD) &&
+        !(local->fd_checked)) {
+        return 1;
+    }
+    return 0;
+}
+
 /* Sets the blocks and size values to fixed values. This is to be called
  * only for dirs. The caller is responsible for checking the type
  */
@@ -4530,6 +4541,7 @@ dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
     int this_call_cnt = 0;
     dht_local_t *local = NULL;
     dht_conf_t *conf = NULL;
+    int ret = 0;
 
     VALIDATE_OR_GOTO(frame, err);
     VALIDATE_OR_GOTO(frame->local, err);
@@ -4538,6 +4550,13 @@ dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
     conf = this->private;
     local = frame->local;
 
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto err;
+        return 0;
+    }
+
     LOCK(&frame->lock);
     {
         if (!xattr || (op_ret == -1)) {
@@ -5205,8 +5224,8 @@ dht_file_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
 
     local->op_errno = op_errno;
 
-    if ((local->fop == GF_FOP_FSETXATTR) && op_ret == -1 &&
-        (op_errno == EBADF) && !(local->fd_checked)) {
+    if ((local->fop == GF_FOP_FSETXATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -5930,8 +5949,8 @@ dht_file_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
 
     local->op_errno = op_errno;
 
-    if ((local->fop == GF_FOP_FREMOVEXATTR) && (op_ret == -1) &&
-        (op_errno == EBADF) && !(local->fd_checked)) {
+    if ((local->fop == GF_FOP_FREMOVEXATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index e126b3574af..3c992297f0f 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -1236,6 +1236,22 @@ dht_newfile_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
                 struct iatt *preparent, struct iatt *postparent, dict_t *xdata);
 
 int
+dht_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *xdata);
+
+int
+dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+                 int op_errno, dict_t *xattr, dict_t *xdata);
+
+int
+dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                       int32_t op_ret, int32_t op_errno, dict_t *dict,
+                       dict_t *xdata);
+int
+dht_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+                 int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata);
+
+int
 gf_defrag_status_get(dht_conf_t *conf, dict_t *dict);
 
 void
@@ -1531,4 +1547,7 @@ int
 dht_pt_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
               dict_t *xdata);
 
+int32_t
+dht_check_remote_fd_failed_error(dht_local_t *local, int op_ret, int op_errno);
+
 #endif /* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index 9d3e3a3a0b2..8d8170bc0ed 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -365,6 +365,23 @@ dht_check_and_open_fd_on_subvol_complete(int ret, call_frame_t *frame,
 
             break;
 
+        case GF_FOP_FXATTROP:
+            STACK_WIND(frame, dht_common_xattrop_cbk, subvol,
+                       subvol->fops->fxattrop, local->fd,
+                       local->rebalance.flags, local->rebalance.xattr,
+                       local->xattr_req);
+            break;
+
+        case GF_FOP_FGETXATTR:
+            STACK_WIND(frame, dht_getxattr_cbk, subvol, subvol->fops->fgetxattr,
+                       local->fd, local->key, NULL);
+            break;
+
+        case GF_FOP_FINODELK:
+            STACK_WIND(frame, dht_finodelk_cbk, subvol, subvol->fops->finodelk,
+                       local->key, local->fd, local->rebalance.lock_cmd,
+                       &local->rebalance.flock, local->xattr_req);
+            break;
         default:
             gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP,
                    "Unknown FOP on fd (%p) on file %s @ %s", fd,
@@ -428,6 +445,18 @@ handle_err:
             DHT_STACK_UNWIND(fremovexattr, frame, -1, op_errno, NULL);
             break;
 
+        case GF_FOP_FXATTROP:
+            DHT_STACK_UNWIND(fxattrop, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FGETXATTR:
+            DHT_STACK_UNWIND(fgetxattr, frame, -1, op_errno, NULL, NULL);
+            break;
+
+        case GF_FOP_FINODELK:
+            DHT_STACK_UNWIND(finodelk, frame, -1, op_errno, NULL);
+            break;
+
         default:
             gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_UNKNOWN_FOP,
                    "Unknown FOP on fd (%p) on file %s @ %s", fd,
diff --git a/xlators/cluster/dht/src/dht-inode-read.c b/xlators/cluster/dht/src/dht-inode-read.c
index cacfe353272..0c209a52312 100644
--- a/xlators/cluster/dht/src/dht-inode-read.c
+++ b/xlators/cluster/dht/src/dht-inode-read.c
@@ -162,8 +162,8 @@ dht_file_attr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
     local = frame->local;
     prev = cookie;
 
-    if ((local->fop == GF_FOP_FSTAT) && (op_ret == -1) && (op_errno == EBADF) &&
-        !(local->fd_checked)) {
+    if ((local->fop == GF_FOP_FSTAT) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -431,7 +431,7 @@ dht_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
     if (local->call_cnt != 1)
         goto out;
 
-    if (op_ret == -1 && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -703,7 +703,7 @@ dht_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
     if (local->call_cnt != 1)
         goto out;
 
-    if (op_ret == -1 && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -820,7 +820,7 @@ dht_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
 
     local->op_errno = op_errno;
 
-    if (op_ret == -1 && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -1223,6 +1223,13 @@ dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
     if (local->call_cnt != 1)
         goto out;
 
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
     ret = dht_read_iatt_from_xdata(this, xdata, &stbuf);
 
     if ((!op_ret) && (ret)) {
@@ -1535,8 +1542,26 @@ dht_finodelk_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
                  int32_t op_ret, int32_t op_errno, dict_t *xdata)
 
 {
+    dht_local_t *local = NULL;
+    int ret = 0;
+
+    GF_VALIDATE_OR_GOTO("dht", frame, out);
+    GF_VALIDATE_OR_GOTO("dht", this, out);
+    GF_VALIDATE_OR_GOTO("dht", frame->local, out);
+
+    local = frame->local;
+
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
+        ret = dht_check_and_open_fd_on_subvol(this, frame);
+        if (ret)
+            goto out;
+        return 0;
+    }
+
+out:
     dht_lk_inode_unref(frame, op_ret);
     DHT_STACK_UNWIND(finodelk, frame, op_ret, op_errno, xdata);
+
     return 0;
 }
 
@@ -1574,6 +1599,13 @@ dht_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
             if (ret)
                     goto err;
     */
+    local->rebalance.flock = *lock;
+    local->rebalance.lock_cmd = cmd;
+    local->key = gf_strdup(volume);
+
+    if (xdata)
+        local->xattr_req = dict_ref(xdata);
+
     STACK_WIND(frame, dht_finodelk_cbk, lock_subvol,
                lock_subvol->fops->finodelk, volume, fd, cmd, lock, xdata);
 
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
index b26b7058d3e..b6b349d64ed 100644
--- a/xlators/cluster/dht/src/dht-inode-write.c
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -49,7 +49,7 @@ dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
      * We only check once as this could be a valid bad fd error.
      */
 
-    if (op_ret == -1 && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -262,8 +262,8 @@ dht_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
      * We only check once as this could actually be a valid error.
      */
 
-    if ((local->fop == GF_FOP_FTRUNCATE) && (op_ret == -1) &&
-        ((op_errno == EBADF) || (op_errno == EINVAL)) && !(local->fd_checked)) {
+    if ((local->fop == GF_FOP_FTRUNCATE) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -489,7 +489,7 @@ dht_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
      * We only check once as this could actually be a valid error.
      */
 
-    if ((op_ret == -1) && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -666,7 +666,7 @@ dht_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
      * and a lookup updated the cached subvol in the inode ctx.
      * We only check once as this could actually be a valid error.
      */
-    if ((op_ret == -1) && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -838,7 +838,7 @@ dht_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
      * and a lookup updated the cached subvol in the inode ctx.
      * We only check once as this could actually be a valid error.
      */
-    if ((op_ret == -1) && (op_errno == EBADF) && !(local->fd_checked)) {
+    if (dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;
@@ -1005,8 +1005,8 @@ dht_file_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
 
     local->op_errno = op_errno;
 
-    if ((local->fop == GF_FOP_FSETATTR) && (op_ret == -1) &&
-        (op_errno == EBADF) && !(local->fd_checked)) {
+    if ((local->fop == GF_FOP_FSETATTR) &&
+        dht_check_remote_fd_failed_error(local, op_ret, op_errno)) {
         ret = dht_check_and_open_fd_on_subvol(this, frame);
         if (ret)
             goto out;