[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[PATCH] dump-guest-memory: Use BQL to protect dump finalize process

From: Peter Xu
Subject: [PATCH] dump-guest-memory: Use BQL to protect dump finalize process
Date: Tue, 16 Nov 2021 11:22:34 +0800

When finalizing the dump-guest-memory with detached mode, we'll first set dump
status to either FAIL or COMPLETE before doing the cleanup, however right after
the dump status change it's possible that another dump-guest-memory qmp command
is sent so both the main thread and dump thread (which is during cleanup) could
be accessing dump state in paralell.  That's racy.

Fix it by protecting the finalizing phase of dump-guest-memory using BQL as
well.  To do that, we expand the BQL from dump_cleanup() into dump_process(),
so we will take the BQL longer than before.  The critical section must cover
the status switch from ACTIVE->{FAIL|COMPLETE} so as to make sure there's no
race any more.

We can also just introduce a specific mutex to serialize the dump process, but
BQL should be enough for now so far, not to mention vm_start() in dump_cleanup
will need BQL anyway, so maybe easier to just use the same mutex.

Reported-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
 dump/dump.c | 24 ++++++++++++++++++------
 1 file changed, 18 insertions(+), 6 deletions(-)

diff --git a/dump/dump.c b/dump/dump.c
index 662d0a62cd..196b7b8ab9 100644
--- a/dump/dump.c
+++ b/dump/dump.c
@@ -96,13 +96,7 @@ static int dump_cleanup(DumpState *s)
     s->guest_note = NULL;
     if (s->resume) {
-        if (s->detached) {
-            qemu_mutex_lock_iothread();
-        }
-        if (s->detached) {
-            qemu_mutex_unlock_iothread();
-        }
@@ -1873,6 +1867,11 @@ static void dump_process(DumpState *s, Error **errp)
     Error *local_err = NULL;
     DumpQueryResult *result = NULL;
+    /*
+     * When running with detached mode, these operations are not run with BQL.
+     * It's still safe, because it's protected by setting s->state to ACTIVE,
+     * so dump_in_progress() check will block yet another dump-guest-memory.
+     */
     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
 #ifdef TARGET_X86_64
         create_win_dump(s, &local_err);
@@ -1883,6 +1882,15 @@ static void dump_process(DumpState *s, Error **errp)
         create_vmcore(s, &local_err);
+    /*
+     * Serialize the finalizing of dump process using BQL to make sure no
+     * concurrent access to DumpState is allowed.  BQL is also required for
+     * dump_cleanup as vm_start() needs it.
+     */
+    if (s->detached) {
+        qemu_mutex_lock_iothread();
+    }
     /* make sure status is written after written_size updates */
@@ -1898,6 +1906,10 @@ static void dump_process(DumpState *s, Error **errp)
     error_propagate(errp, local_err);
+    if (s->detached) {
+        qemu_mutex_unlock_iothread();
+    }
 static void *dump_thread(void *data)

reply via email to

[Prev in Thread] Current Thread [Next in Thread]