fix some segmentation faults and migration issues

When qemu_signal_init() fails in qemu_init_main_loop(), we return without setting an error. Its callers crash then when they try to report the error with error_report_err().

To avoid such segmentation fault, add a new Error parameter to make the call trace to propagate the err to the final caller.

Fixes 2f78e491d7b46542158ce0b8132ee4e05bc0ade4

Cc Markus Armbruster <[email protected]>

Cc Fam Zheng <[email protected]>

Signed-off-by Fei Li <[email protected]>

Reviewed-by Fam Zheng <[email protected]>

Reviewed-by Markus Armbruster <[email protected]>

---
 util/main-loop.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

util/main-loop.c +4 -4

index affe0403c5..443cb4cfe8 100644
@@ -71,7 +71,7 @@ static void sigfd_handler(void *opaque)
     }
 }
 
-static int qemu_signal_init(void)
+static int qemu_signal_init(Error **errp)
 {
     int sigfd;
     sigset_t set;

@@ -96,7 +96,7 @@ static int qemu_signal_init(void)
     sigdelset(&set, SIG_IPI);
     sigfd = qemu_signalfd(&set);
     if (sigfd == -1) {
-        fprintf(stderr, "failed to create signalfd\n");
+        error_setg_errno(errp, errno, "failed to create signalfd");
         return -errno;
     }
 

@@ -109,7 +109,7 @@ static int qemu_signal_init(void)
 
 #else /* _WIN32 */
 
-static int qemu_signal_init(void)
+static int qemu_signal_init(Error **errp)
 {
     return 0;
 }

@@ -148,7 +148,7 @@ int qemu_init_main_loop(Error **errp)
 
     init_clocks(qemu_timer_notify_cb);
 
-    ret = qemu_signal_init();
+    ret = qemu_signal_init(errp);
     if (ret) {
         return ret;
     }

You neglected to cc the maintainer. I'm doing that for you now. Cc'ing maintainers is important to maximize your chances at getting your patches picked up. Use scripts/get_maintainer.pl to find them.

Quoted 57 lines · Fei Li

When qemu_signal_init() fails in qemu_init_main_loop(), we return without setting an error. Its callers crash then when they try to report the error with error_report_err().

To avoid such segmentation fault, add a new Error parameter to make the call trace to propagate the err to the final caller.

Fixes 2f78e491d7b46542158ce0b8132ee4e05bc0ade4

Cc Markus Armbruster <[email protected]>

Cc Fam Zheng <[email protected]>

Signed-off-by Fei Li <[email protected]>

Reviewed-by Fam Zheng <[email protected]>

Reviewed-by Markus Armbruster <[email protected]>

---
  util/main-loop.c | 8 ++++----
  1 file changed, 4 insertions(+), 4 deletions(-)

util/main-loop.c +4 -4

index affe0403c5..443cb4cfe8 100644
@@ -71,7 +71,7 @@ static void sigfd_handler(void *opaque)
      }
  }
 
-static int qemu_signal_init(void)
+static int qemu_signal_init(Error **errp)
  {
      int sigfd;
      sigset_t set;

@@ -96,7 +96,7 @@ static int qemu_signal_init(void)
      sigdelset(&set, SIG_IPI);
      sigfd = qemu_signalfd(&set);
      if (sigfd == -1) {
-        fprintf(stderr, "failed to create signalfd\n");
+        error_setg_errno(errp, errno, "failed to create signalfd");
          return -errno;
      }
 

@@ -109,7 +109,7 @@ static int qemu_signal_init(void)
 
  #else /* _WIN32 */
 
-static int qemu_signal_init(void)
+static int qemu_signal_init(Error **errp)
  {
      return 0;
  }

@@ -148,7 +148,7 @@ int qemu_init_main_loop(Error **errp)
 
      init_clocks(qemu_timer_notify_cb);
 
-    ret = qemu_signal_init();
+    ret = qemu_signal_init(errp);
      if (ret) {
          return ret;
      }

In our current code, when multifd is used during migration, if there is an error before the destination receives all new channels, the source keeps running, however the destination does not exit but keeps waiting until the source is killed deliberately.

Fix this by dumping the specific error and let users decide whether to quit from the destination side when failing to receive packet via some channel.

Signed-off-by Fei Li <[email protected]>

Reviewed-by Peter Xu <[email protected]>

---
 migration/channel.c   | 11 ++++++-----
 migration/migration.c |  9 +++++++--
 migration/migration.h |  2 +-
 migration/ram.c       |  7 ++++++-
 migration/ram.h       |  2 +-
 5 files changed, 21 insertions(+), 10 deletions(-)

migration/channel.c +6 -5

index 33e0e9b82f..20e4c8e2dc 100644
@@ -30,6 +30,7 @@
 void migration_channel_process_incoming(QIOChannel *ioc)
 {
     MigrationState *s = migrate_get_current();
+    Error *local_err = NULL;
 
     trace_migration_set_incoming_channel(
         ioc, object_get_typename(OBJECT(ioc)));

@@ -38,13 +39,13 @@ void migration_channel_process_incoming(QIOChannel *ioc)
         *s->parameters.tls_creds &&
         !object_dynamic_cast(OBJECT(ioc),
                              TYPE_QIO_CHANNEL_TLS)) {
-        Error *local_err = NULL;
         migration_tls_channel_process_incoming(s, ioc, &local_err);
-        if (local_err) {
-            error_report_err(local_err);
-        }
     } else {
-        migration_ioc_process_incoming(ioc);
+        migration_ioc_process_incoming(ioc, &local_err);
+    }
+
+    if (local_err) {
+        error_report_err(local_err);
     }
 }
 

migration/migration.c +7 -2

index 49ffb9997a..72106bddf0 100644
@@ -541,7 +541,7 @@ void migration_fd_process_incoming(QEMUFile *f)
     migration_incoming_process();
 }
 
-void migration_ioc_process_incoming(QIOChannel *ioc)
+void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
 {
     MigrationIncomingState *mis = migration_incoming_get_current();
     bool start_migration;

@@ -563,9 +563,14 @@ void migration_ioc_process_incoming(QIOChannel *ioc)
          */
         start_migration = !migrate_use_multifd();
     } else {
+        Error *local_err = NULL;
         /* Multiple connections */
         assert(migrate_use_multifd());
-        start_migration = multifd_recv_new_channel(ioc);
+        start_migration = multifd_recv_new_channel(ioc, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
     }
 
     if (start_migration) {

migration/migration.h +1 -1

index e413d4d8b6..02b7304610 100644
@@ -229,7 +229,7 @@ struct MigrationState
 void migrate_set_state(int *state, int old_state, int new_state);
 
 void migration_fd_process_incoming(QEMUFile *f);
-void migration_ioc_process_incoming(QIOChannel *ioc);
+void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
 void migration_incoming_process(void);
 
 bool  migration_has_all_channels(void);

migration/ram.c +1 -1

index 7e7deec4d8..e13b9349d0 100644
@@ -1323,7 +1323,7 @@ bool multifd_recv_all_channels_created(void)
 }
 
 /* Return true if multifd is ready for the migration, otherwise false */
-bool multifd_recv_new_channel(QIOChannel *ioc)
+bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
 {
     MultiFDRecvParams *p;
     Error *local_err = NULL;

migration/ram.c +1 -0

         multifd_recv_terminate_threads(local_err);
         return false;
     }

@@ -1340,6 +1344,7 @@ bool multifd_recv_new_channel(QIOChannel *ioc)
         error_setg(&local_err, "multifd: received id '%d' already setup'",
                    id);
         multifd_recv_terminate_threads(local_err);
+        error_propagate(errp, local_err);
         return false;
     }
     p->c = ioc;

migration/ram.h +1 -1

index 83ff1bc11a..046d3074be 100644
@@ -47,7 +47,7 @@ int multifd_save_cleanup(Error **errp);
 int multifd_load_setup(void);
 int multifd_load_cleanup(Error **errp);
 bool multifd_recv_all_channels_created(void);
-bool multifd_recv_new_channel(QIOChannel *ioc);
+bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp);
 
 uint64_t ram_pagesize_summary(void);
 int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len);

Hi Fei,

Quoted 107 lines · Fei Li

In our current code, when multifd is used during migration, if there is an error before the destination receives all new channels, the source keeps running, however the destination does not exit but keeps waiting until the source is killed deliberately.

Fix this by dumping the specific error and let users decide whether to quit from the destination side when failing to receive packet via some channel.

Signed-off-by Fei Li <[email protected]>

Reviewed-by Peter Xu <[email protected]>

---
  migration/channel.c   | 11 ++++++-----
  migration/migration.c |  9 +++++++--
  migration/migration.h |  2 +-
  migration/ram.c       |  7 ++++++-
  migration/ram.h       |  2 +-
  5 files changed, 21 insertions(+), 10 deletions(-)

migration/channel.c +6 -5

index 33e0e9b82f..20e4c8e2dc 100644
@@ -30,6 +30,7 @@
  void migration_channel_process_incoming(QIOChannel *ioc)
  {
      MigrationState *s = migrate_get_current();
+    Error *local_err = NULL;
 
      trace_migration_set_incoming_channel(
          ioc, object_get_typename(OBJECT(ioc)));

@@ -38,13 +39,13 @@ void migration_channel_process_incoming(QIOChannel *ioc)
          *s->parameters.tls_creds &&
          !object_dynamic_cast(OBJECT(ioc),
                               TYPE_QIO_CHANNEL_TLS)) {
-        Error *local_err = NULL;
          migration_tls_channel_process_incoming(s, ioc, &local_err);
-        if (local_err) {
-            error_report_err(local_err);
-        }
      } else {
-        migration_ioc_process_incoming(ioc);
+        migration_ioc_process_incoming(ioc, &local_err);
+    }
+
+    if (local_err) {
+        error_report_err(local_err);
      }
  }
 

migration/migration.c +7 -2

index 49ffb9997a..72106bddf0 100644
@@ -541,7 +541,7 @@ void migration_fd_process_incoming(QEMUFile *f)
      migration_incoming_process();
  }
 
-void migration_ioc_process_incoming(QIOChannel *ioc)
+void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
  {
      MigrationIncomingState *mis = migration_incoming_get_current();
      bool start_migration;

@@ -563,9 +563,14 @@ void migration_ioc_process_incoming(QIOChannel *ioc)
           */
          start_migration = !migrate_use_multifd();
      } else {
+        Error *local_err = NULL;
          /* Multiple connections */
          assert(migrate_use_multifd());
-        start_migration = multifd_recv_new_channel(ioc);
+        start_migration = multifd_recv_new_channel(ioc, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
      }
 
      if (start_migration) {

migration/migration.h +1 -1

index e413d4d8b6..02b7304610 100644
@@ -229,7 +229,7 @@ struct MigrationState
  void migrate_set_state(int *state, int old_state, int new_state);
 
  void migration_fd_process_incoming(QEMUFile *f);
-void migration_ioc_process_incoming(QIOChannel *ioc);
+void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
  void migration_incoming_process(void);
 
  bool  migration_has_all_channels(void);

migration/ram.c +1 -1

index 7e7deec4d8..e13b9349d0 100644
@@ -1323,7 +1323,7 @@ bool multifd_recv_all_channels_created(void)
  }
 
  /* Return true if multifd is ready for the migration, otherwise false */
-bool multifd_recv_new_channel(QIOChannel *ioc)
+bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
  {
      MultiFDRecvParams *p;
      Error *local_err = NULL;

@@ -1331,6 +1331,10 @@ bool multifd_recv_new_channel(QIOChannel *ioc)
 

migration/ram.c bool multifd_recv_new_channel(QIOChannel *ioc) · line 1337

Quoted 6 lines

migration/ram.c +4 -0

      id = multifd_recv_initial_packet(ioc, &local_err);
      if (id < 0) {
+        error_propagate_prepend(errp, local_err,
+                                "failed to receive packet"
+                                " via multifd channel %d: ",
+                                multifd_recv_state->count);

Shouldn't we use atomic_read(&multifd_recv_state->count) here?

Patch looks good otherwise.

Regards,

Phil.

Quoted 25 lines

migration/ram.c +1 -0

          multifd_recv_terminate_threads(local_err);
          return false;
      }

@@ -1340,6 +1344,7 @@ bool multifd_recv_new_channel(QIOChannel *ioc)
          error_setg(&local_err, "multifd: received id '%d' already setup'",
                     id);
          multifd_recv_terminate_threads(local_err);
+        error_propagate(errp, local_err);
          return false;
      }
      p->c = ioc;

migration/ram.h +1 -1

index 83ff1bc11a..046d3074be 100644
@@ -47,7 +47,7 @@ int multifd_save_cleanup(Error **errp);
  int multifd_load_setup(void);
  int multifd_load_cleanup(Error **errp);
  bool multifd_recv_all_channels_created(void);
-bool multifd_recv_new_channel(QIOChannel *ioc);
+bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp);
 
  uint64_t ram_pagesize_summary(void);
  int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len);

Quoted 103 lines · Philippe Mathieu-Daudé

Hi Fei,

Quoted 100 lines · Fei Li

In our current code, when multifd is used during migration, if there is an error before the destination receives all new channels, the source keeps running, however the destination does not exit but keeps waiting until the source is killed deliberately.

Fix this by dumping the specific error and let users decide whether to quit from the destination side when failing to receive packet via some channel.

Signed-off-by Fei Li <[email protected]>

Reviewed-by Peter Xu <[email protected]>

---
  migration/channel.c   | 11 ++++++-----
  migration/migration.c |  9 +++++++--
  migration/migration.h |  2 +-
  migration/ram.c       |  7 ++++++-
  migration/ram.h       |  2 +-

  5 files changed, 21 insertions(+), 10 deletions(-)

migration/channel.c +1 -0

index 33e0e9b82f..20e4c8e2dc 100644
@@ -30,6 +30,7 @@
   void migration_channel_process_incoming(QIOChannel *ioc)
   {
       MigrationState *s = migrate_get_current();
+    Error *local_err = NULL;

        trace_migration_set_incoming_channel(
          ioc, object_get_typename(OBJECT(ioc)));

migration/channel.c +5 -5

@@ -38,13 +39,13 @@ void 
 migration_channel_process_incoming(QIOChannel *ioc)
           *s->parameters.tls_creds &&
           !object_dynamic_cast(OBJECT(ioc),
                                TYPE_QIO_CHANNEL_TLS)) {
-        Error *local_err = NULL;
           migration_tls_channel_process_incoming(s, ioc, &local_err);
-        if (local_err) {
-            error_report_err(local_err);
-        }
       } else {
-        migration_ioc_process_incoming(ioc);
+        migration_ioc_process_incoming(ioc, &local_err);
+    }
+
+    if (local_err) {
+        error_report_err(local_err);
       }
   }

  diff --git a/migration/migration.c b/migration/migration.c index 49ffb9997a..72106bddf0 100644

migration/migration.c +7 -1

@@ -541,7 +541,7 @@ void migration_fd_process_incoming(QEMUFile *f)
       migration_incoming_process();
   }
   -void migration_ioc_process_incoming(QIOChannel *ioc)
+void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
   {
       MigrationIncomingState *mis = migration_incoming_get_current();
       bool start_migration;

@@ -563,9 +563,14 @@ void migration_ioc_process_incoming(QIOChannel 
 *ioc)
            */
           start_migration = !migrate_use_multifd();
       } else {
+        Error *local_err = NULL;
           /* Multiple connections */
           assert(migrate_use_multifd());
-        start_migration = multifd_recv_new_channel(ioc);
+        start_migration = multifd_recv_new_channel(ioc, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
       }
         if (start_migration) {

migration/migration.h +1 -1

index e413d4d8b6..02b7304610 100644
@@ -229,7 +229,7 @@ struct MigrationState
   void migrate_set_state(int *state, int old_state, int new_state);
     void migration_fd_process_incoming(QEMUFile *f);
-void migration_ioc_process_incoming(QIOChannel *ioc);
+void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
   void migration_incoming_process(void);
     bool  migration_has_all_channels(void);

migration/ram.c +1 -1

index 7e7deec4d8..e13b9349d0 100644
@@ -1323,7 +1323,7 @@ bool multifd_recv_all_channels_created(void)
   }
     /* Return true if multifd is ready for the migration, otherwise false */
-bool multifd_recv_new_channel(QIOChannel *ioc)
+bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp)
   {
       MultiFDRecvParams *p;
       Error *local_err = NULL;

Quoted 8 lines

migration/ram.c bool multifd_recv_new_channel(QIOChannel *ioc) · line 1336

Quoted 7 lines

migration/ram.c +4 -0

@@ -1331,6 +1331,10 @@ bool multifd_recv_new_channel(QIOChannel *ioc)
         id = multifd_recv_initial_packet(ioc, &local_err);
       if (id < 0) {
+        error_propagate_prepend(errp, local_err,
+                                "failed to receive packet"
+                                " via multifd channel %d: ",
+ multifd_recv_state->count);

Shouldn't we use atomic_read(&multifd_recv_state->count) here?

Right, will update this in next version. Thanks for pointing it out. :) BTW, should we do the same update for the below sentence:
return multifd_recv_state->count == migrate_multifd_channels();?

Have a nice day
Fei

Quoted 32 lines

Patch looks good otherwise.

Regards,

Phil.

Quoted 24 lines

migration/ram.c +1 -0

 multifd_recv_terminate_threads(local_err);
           return false;
       }

@@ -1340,6 +1344,7 @@ bool multifd_recv_new_channel(QIOChannel *ioc)
           error_setg(&local_err, "multifd: received id '%d' already setup'",
                      id);
           multifd_recv_terminate_threads(local_err);
+        error_propagate(errp, local_err);
           return false;
       }
       p->c = ioc;

migration/ram.h +1 -1

index 83ff1bc11a..046d3074be 100644
@@ -47,7 +47,7 @@ int multifd_save_cleanup(Error **errp);
   int multifd_load_setup(void);
   int multifd_load_cleanup(Error **errp);
   bool multifd_recv_all_channels_created(void);
-bool multifd_recv_new_channel(QIOChannel *ioc);
+bool multifd_recv_new_channel(QIOChannel *ioc, Error **errp);
     uint64_t ram_pagesize_summary(void);
   int ram_save_queue_pages(const char *rbname, ram_addr_t start, ram_addr_t len);
 

Always call migrate_set_error() to set the error state without relying on whether multifd_save_cleanup() succeeds. As the passed &local_err is never used in multifd_save_cleanup(), remove it. And make the function be: void multifd_save_cleanup(void).

Signed-off-by Fei Li <[email protected]>

Reviewed-by Juan Quintela <[email protected]>

---
 migration/migration.c |  5 +----
 migration/ram.c       | 11 ++++-------
 migration/ram.h       |  2 +-
 3 files changed, 6 insertions(+), 12 deletions(-)

migration/migration.c +1 -4

index 72106bddf0..0537fc0c26 100644
@@ -1386,7 +1386,6 @@ static void migrate_fd_cleanup(void *opaque)
     qemu_savevm_state_cleanup();
 
     if (s->to_dst_file) {
-        Error *local_err = NULL;
         QEMUFile *tmp;
 
         trace_migrate_fd_cleanup();

@@ -1397,9 +1396,7 @@ static void migrate_fd_cleanup(void *opaque)
         }
         qemu_mutex_lock_iothread();
 
-        if (multifd_save_cleanup(&local_err) != 0) {
-            error_report_err(local_err);
-        }
+        multifd_save_cleanup();
         qemu_mutex_lock(&s->qemu_file_lock);
         tmp = s->to_dst_file;
         s->to_dst_file = NULL;

migration/ram.c +4 -7

index e13b9349d0..c44cb6f56d 100644
@@ -917,13 +917,12 @@ static void multifd_send_terminate_threads(Error *err)
     }
 }
 
-int multifd_save_cleanup(Error **errp)
+void multifd_save_cleanup(void)
 {
     int i;
-    int ret = 0;
 
     if (!migrate_use_multifd()) {
-        return 0;
+        return;
     }
     multifd_send_terminate_threads(NULL);
     for (i = 0; i < migrate_multifd_channels(); i++) {

@@ -953,7 +952,6 @@ int multifd_save_cleanup(Error **errp)
     multifd_send_state->pages = NULL;
     g_free(multifd_send_state);
     multifd_send_state = NULL;
-    return ret;
 }
 
 static void multifd_send_sync_main(void)

@@ -1071,9 +1069,8 @@ static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
     Error *local_err = NULL;
 
     if (qio_task_propagate_error(task, &local_err)) {
-        if (multifd_save_cleanup(&local_err) != 0) {
-            migrate_set_error(migrate_get_current(), local_err);
-        }
+        migrate_set_error(migrate_get_current(), local_err);
+        multifd_save_cleanup();
     } else {
         p->c = QIO_CHANNEL(sioc);
         qio_channel_set_delay(p->c, false);

migration/ram.h +1 -1

index 046d3074be..936177b3e9 100644
@@ -43,7 +43,7 @@ uint64_t ram_bytes_remaining(void);
 uint64_t ram_bytes_total(void);
 
 int multifd_save_setup(void);
-int multifd_save_cleanup(Error **errp);
+void multifd_save_cleanup(void);
 int multifd_load_setup(void);
 int multifd_load_cleanup(Error **errp);
 bool multifd_recv_all_channels_created(void);

Quoted 4 lines · Fei Li

Always call migrate_set_error() to set the error state without relying on whether multifd_save_cleanup() succeeds. As the passed &local_err is never used in multifd_save_cleanup(), remove it. And make the function be: void multifd_save_cleanup(void).

Reading this after your patch 1/5, maybe multifd_save_cleanup() lacks

error reporting on failure. So the actual prototype is correct, we just need to use **errp in that function (like reporting invalid thread instead of calling qemu_thread_join()).

Quoted 86 lines

Signed-off-by Fei Li <[email protected]>

Reviewed-by Juan Quintela <[email protected]>

---
 migration/migration.c |  5 +----
 migration/ram.c       | 11 ++++-------
 migration/ram.h       |  2 +-
 3 files changed, 6 insertions(+), 12 deletions(-)

migration/migration.c +1 -4

index 72106bddf0..0537fc0c26 100644
@@ -1386,7 +1386,6 @@ static void migrate_fd_cleanup(void *opaque)
     qemu_savevm_state_cleanup();
 
     if (s->to_dst_file) {
-        Error *local_err = NULL;
         QEMUFile *tmp;
 
         trace_migrate_fd_cleanup();

@@ -1397,9 +1396,7 @@ static void migrate_fd_cleanup(void *opaque)
         }
         qemu_mutex_lock_iothread();
 
-        if (multifd_save_cleanup(&local_err) != 0) {
-            error_report_err(local_err);
-        }
+        multifd_save_cleanup();
         qemu_mutex_lock(&s->qemu_file_lock);
         tmp = s->to_dst_file;
         s->to_dst_file = NULL;

migration/ram.c +4 -7

index e13b9349d0..c44cb6f56d 100644
@@ -917,13 +917,12 @@ static void multifd_send_terminate_threads(Error *err)
     }
 }
 
-int multifd_save_cleanup(Error **errp)
+void multifd_save_cleanup(void)
 {
     int i;
-    int ret = 0;
 
     if (!migrate_use_multifd()) {
-        return 0;
+        return;
     }
     multifd_send_terminate_threads(NULL);
     for (i = 0; i < migrate_multifd_channels(); i++) {

@@ -953,7 +952,6 @@ int multifd_save_cleanup(Error **errp)
     multifd_send_state->pages = NULL;
     g_free(multifd_send_state);
     multifd_send_state = NULL;
-    return ret;
 }
 
 static void multifd_send_sync_main(void)

@@ -1071,9 +1069,8 @@ static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
     Error *local_err = NULL;
 
     if (qio_task_propagate_error(task, &local_err)) {
-        if (multifd_save_cleanup(&local_err) != 0) {
-            migrate_set_error(migrate_get_current(), local_err);
-        }
+        migrate_set_error(migrate_get_current(), local_err);
+        multifd_save_cleanup();
     } else {
         p->c = QIO_CHANNEL(sioc);
         qio_channel_set_delay(p->c, false);

migration/ram.h +1 -1

index 046d3074be..936177b3e9 100644
@@ -43,7 +43,7 @@ uint64_t ram_bytes_remaining(void);
 uint64_t ram_bytes_total(void);
 
 int multifd_save_setup(void);
-int multifd_save_cleanup(Error **errp);
+void multifd_save_cleanup(void);
 int multifd_load_setup(void);
 int multifd_load_cleanup(Error **errp);
 bool multifd_recv_all_channels_created(void);

Quoted 4 lines · Fei Li

Always call migrate_set_error() to set the error state without relying on whether multifd_save_cleanup() succeeds. As the passed &local_err is never used in multifd_save_cleanup(), remove it. And make the function be: void multifd_save_cleanup(void).

Reading this after your patch 1/5, maybe multifd_save_cleanup() lacks error reporting on failure. So the actual prototype is correct, we just need to use **errp in that function (like reporting invalid thread instead of calling qemu_thread_join()).

Quoted 85 lines

Signed-off-by Fei Li <[email protected]>

Reviewed-by Juan Quintela <[email protected]>

---
  migration/migration.c |  5 +----
  migration/ram.c       | 11 ++++-------
  migration/ram.h       |  2 +-
  3 files changed, 6 insertions(+), 12 deletions(-)

migration/migration.c +1 -4

index 72106bddf0..0537fc0c26 100644
@@ -1386,7 +1386,6 @@ static void migrate_fd_cleanup(void *opaque)
      qemu_savevm_state_cleanup();
 
      if (s->to_dst_file) {
-        Error *local_err = NULL;
          QEMUFile *tmp;
 
          trace_migrate_fd_cleanup();

@@ -1397,9 +1396,7 @@ static void migrate_fd_cleanup(void *opaque)
          }
          qemu_mutex_lock_iothread();
 
-        if (multifd_save_cleanup(&local_err) != 0) {
-            error_report_err(local_err);
-        }
+        multifd_save_cleanup();
          qemu_mutex_lock(&s->qemu_file_lock);
          tmp = s->to_dst_file;
          s->to_dst_file = NULL;

migration/ram.c +4 -7

index e13b9349d0..c44cb6f56d 100644
@@ -917,13 +917,12 @@ static void multifd_send_terminate_threads(Error *err)
      }
  }
 
-int multifd_save_cleanup(Error **errp)
+void multifd_save_cleanup(void)
  {
      int i;
-    int ret = 0;
 
      if (!migrate_use_multifd()) {
-        return 0;
+        return;
      }
      multifd_send_terminate_threads(NULL);
      for (i = 0; i < migrate_multifd_channels(); i++) {

@@ -953,7 +952,6 @@ int multifd_save_cleanup(Error **errp)
      multifd_send_state->pages = NULL;
      g_free(multifd_send_state);
      multifd_send_state = NULL;
-    return ret;
  }
 
  static void multifd_send_sync_main(void)

@@ -1071,9 +1069,8 @@ static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque)
      Error *local_err = NULL;
 
      if (qio_task_propagate_error(task, &local_err)) {
-        if (multifd_save_cleanup(&local_err) != 0) {
-            migrate_set_error(migrate_get_current(), local_err);
-        }
+        migrate_set_error(migrate_get_current(), local_err);
+        multifd_save_cleanup();
      } else {
          p->c = QIO_CHANNEL(sioc);
          qio_channel_set_delay(p->c, false);

migration/ram.h +1 -1

index 046d3074be..936177b3e9 100644
@@ -43,7 +43,7 @@ uint64_t ram_bytes_remaining(void);
  uint64_t ram_bytes_total(void);
 
  int multifd_save_setup(void);
-int multifd_save_cleanup(Error **errp);
+void multifd_save_cleanup(void);
  int multifd_load_setup(void);
  int multifd_load_cleanup(Error **errp);
  bool multifd_recv_all_channels_created(void);

Call postcopy_ram_incoming_cleanup() to do the cleanup when postcopy_ram_enable_notify fails. Besides, report the error message when qemu_ram_foreach_migratable_block() fails.

Cc Dr. David Alan Gilbert <[email protected]>

Signed-off-by Fei Li <[email protected]>

---
 migration/postcopy-ram.c | 1 +
 migration/savevm.c       | 1 +
 2 files changed, 2 insertions(+)

migration/postcopy-ram.c +1 -0

index e5c02a32c5..fa09dba534 100644
@@ -1117,6 +1117,7 @@ int postcopy_ram_enable_notify(MigrationIncomingState *mis)
 
     /* Mark so that we get notified of accesses to unwritten areas */
     if (qemu_ram_foreach_migratable_block(ram_block_enable_notify, mis)) {
+        error_report("ram_block_enable_notify failed");
         return -1;
     }
 

migration/savevm.c +1 -0

index 9e45fb4f3f..d784e8aa40 100644
@@ -1729,6 +1729,7 @@ static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis)
      */
     if (migrate_postcopy_ram()) {
         if (postcopy_ram_enable_notify(mis)) {
+            postcopy_ram_incoming_cleanup(mis);
             return -1;
         }
     }

Signature · 1 line

2.13.7

These five patches almost get the Reviewed-by and are extracted from previous "[PATCH RFC v7 0/9] qemu_thread_create: propagate errors to callers to check." The mentioned patch series have waited on one multifd issue for a while and still needs a further discussion.

Thus separate(send) these five almost-done patches and hope they can be merged for the next tag. Thanks for the review. :)

v2:
- Update the commit message for patch 1/5, and get one more Reviewed-by.
- Get one Reviewed-by for patch 3/5.

Fei Li (5):
Fix segmentation fault when qemu_signal_init fails
qemu_thread_join: fix segmentation fault
migration: fix the multifd code when receiving less channels
migration: remove unused &local_err parameter in multifd_save_cleanup
migration: add more error handling for postcopy_ram_enable_notify

Quoted 7 lines · Fei Li

These five patches almost get the Reviewed-by and are extracted from previous "[PATCH RFC v7 0/9] qemu_thread_create: propagate errors to callers to check."  The mentioned patch series have waited on one multifd issue for a while and still needs a further discussion.

Thus separate(send) these five almost-done patches and hope they can be merged for the next tag. Thanks for the review. :)

How likely are any of these crashers to affect an end user?

Are any of them regressions over 3.0?

I'm trying to gauge if any of this is serious enough to warrant a -rc4, or if we are okay just documenting them as known corner-case bugs and deferring the fix to 4.0 and qemu-stable.

The fact that the series is still titled RFC is also an argument in favor of deferral.

Quoted 14 lines

v2:
- Update the commit message for patch 1/5, and get one more Reviewed-by.
- Get one Reviewed-by for patch 3/5.

Fei Li (5):
   Fix segmentation fault when qemu_signal_init fails
   qemu_thread_join: fix segmentation fault
   migration: fix the multifd code when receiving less channels
   migration: remove unused &local_err parameter in multifd_save_cleanup
   migration: add more error handling for postcopy_ram_enable_notify

Quoted 6 lines · Eric Blake

Quoted 5 lines · Fei Li

These five patches almost get the Reviewed-by and are extracted from previous "[PATCH RFC v7 0/9] qemu_thread_create: propagate errors to callers to check."  The mentioned patch series have waited on one multifd issue for a while and still needs a further discussion.

Quoted 4 lines

Quoted 2 lines

Thus separate(send) these five almost-done patches and hope they can be merged for the next tag. Thanks for the review. :)

How likely are any of these crashers to affect an end user?

IMHO, they are not easily triggered.

Quoted 1 line

Are any of them regressions over 3.0?

I do not think so.

Quoted 3 lines

I'm trying to gauge if any of this is serious enough to warrant a -rc4, or if we are okay just documenting them as known corner-case bugs and deferring the fix to 4.0 and qemu-stable.

Emm, actually not so emergency to be included in -rc4.
And I think it is ok to wait for maintainers to do the pick for the appropriate release. BTW, why 4.0, but not 3.2 or 3.3 (I mean 3.minor version)?