iscsi: fix races between task completion and abort

This patch fixes two main issues with block/iscsi.c:

1) iscsi_task_mgmt_abort_task_async calls iscsi_scsi_task_cancel which
was also directly called in iscsi_aio_cancel

2) a race between task completion and task abortion could happen cause
the scsi_free_scsi_task were done before iscsi_schedule_bh has finished.
To fix this, all the freeing of IscsiTasks and releasing of the AIOCBs
is centralized in iscsi_bh_cb, independent of whether the SCSI command
has completed or was cancelled.

3) iscsi_aio_cancel was not synchronously waiting for the end of the
command.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit 1bd075f29ea6d11853475c7c42734595720c3ac6)

Conflicts:

	block/iscsi.c

Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>

1 files changed, 28 insertions, 24 deletions

diff --git a/block/iscsi.c b/block/iscsi.c
index 4584d54eb..c3d8db8b4 100644
--- a/block/iscsi.c
+++ b/block/iscsi.c
@@ -76,12 +76,20 @@ iscsi_bh_cb(void *p)
         acb->common.cb(acb->common.opaque, acb->status);
     }
 
+    if (acb->task != NULL) {
+        scsi_free_scsi_task(acb->task);
+        acb->task = NULL;
+    }
+
     qemu_aio_release(acb);
 }
 
 static void
 iscsi_schedule_bh(IscsiAIOCB *acb)
 {
+    if (acb->bh) {
+        return;
+    }
     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
     qemu_bh_schedule(acb->bh);
 }
@@ -91,6 +99,10 @@ static void
 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
                     void *private_data)
 {
+    IscsiAIOCB *acb = private_data;
+
+    acb->status = -ECANCELED;
+    iscsi_schedule_bh(acb);
 }
 
 static void
@@ -99,15 +111,19 @@ iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
     IscsiLun *iscsilun = acb->iscsilun;
 
-    acb->common.cb(acb->common.opaque, -ECANCELED);
+    if (acb->status != -EINPROGRESS) {
+        return;
+    }
+
     acb->canceled = 1;
 
     /* send a task mgmt call to the target to cancel the task on the target */
     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
-                                     iscsi_abort_task_cb, NULL);
+                                     iscsi_abort_task_cb, acb);
 
-    /* then also cancel the task locally in libiscsi */
-    iscsi_scsi_task_cancel(iscsilun->iscsi, acb->task);
+    while (acb->status == -EINPROGRESS) {
+        qemu_aio_wait();
+    }
 }
 
 static AIOPool iscsi_aio_pool = {
@@ -185,9 +201,6 @@ iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
     g_free(acb->buf);
 
     if (acb->canceled != 0) {
-        qemu_aio_release(acb);
-        scsi_free_scsi_task(acb->task);
-        acb->task = NULL;
         return;
     }
 
@@ -199,8 +212,6 @@ iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
     }
 
     iscsi_schedule_bh(acb);
-    scsi_free_scsi_task(acb->task);
-    acb->task = NULL;
 }
 
 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
@@ -229,6 +240,8 @@ iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
     acb->qiov     = qiov;
 
     acb->canceled   = 0;
+    acb->bh         = NULL;
+    acb->status     = -EINPROGRESS;
 
     /* XXX we should pass the iovec to write16 to avoid the extra copy */
     /* this will allow us to get rid of 'buf' completely */
@@ -287,9 +300,6 @@ iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 
     if (acb->canceled != 0) {
-        qemu_aio_release(acb);
-        scsi_free_scsi_task(acb->task);
-        acb->task = NULL;
         return;
     }
 
@@ -301,8 +311,6 @@ iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
     }
 
     iscsi_schedule_bh(acb);
-    scsi_free_scsi_task(acb->task);
-    acb->task = NULL;
 }
 
 static BlockDriverAIOCB *
@@ -328,6 +336,8 @@ iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
     acb->qiov     = qiov;
 
     acb->canceled    = 0;
+    acb->bh          = NULL;
+    acb->status      = -EINPROGRESS;
     acb->read_size   = qemu_read_size;
     acb->buf         = NULL;
 
@@ -403,9 +413,6 @@ iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
     IscsiAIOCB *acb = opaque;
 
     if (acb->canceled != 0) {
-        qemu_aio_release(acb);
-        scsi_free_scsi_task(acb->task);
-        acb->task = NULL;
         return;
     }
 
@@ -417,8 +424,6 @@ iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
     }
 
     iscsi_schedule_bh(acb);
-    scsi_free_scsi_task(acb->task);
-    acb->task = NULL;
 }
 
 static BlockDriverAIOCB *
@@ -433,6 +438,8 @@ iscsi_aio_flush(BlockDriverState *bs,
 
     acb->iscsilun = iscsilun;
     acb->canceled   = 0;
+    acb->bh         = NULL;
+    acb->status     = -EINPROGRESS;
 
     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
                                          0, 0, 0, 0,
@@ -457,9 +464,6 @@ iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
     IscsiAIOCB *acb = opaque;
 
     if (acb->canceled != 0) {
-        qemu_aio_release(acb);
-        scsi_free_scsi_task(acb->task);
-        acb->task = NULL;
         return;
     }
 
@@ -471,8 +475,6 @@ iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
     }
 
     iscsi_schedule_bh(acb);
-    scsi_free_scsi_task(acb->task);
-    acb->task = NULL;
 }
 
 static BlockDriverAIOCB *
@@ -489,6 +491,8 @@ iscsi_aio_discard(BlockDriverState *bs,
 
     acb->iscsilun = iscsilun;
     acb->canceled   = 0;
+    acb->bh         = NULL;
+    acb->status     = -EINPROGRESS;
 
     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;