[CRIU] [crtools-bot for Pavel Emelyanov ] dump: Try to seize task tree several times on error

[Cyrill Gorcunov]

The commit is pushed to "master" and will appear on git://github.com/cyrillos/crtools.git
------>
commit a1fe2c58a9a0a17e24ab62ecdcfceb22c8f55065
Author: Pavel Emelyanov <xemul at parallels.com>
Date:   Thu Mar 1 19:11:29 2012 +0400

    dump: Try to seize task tree several times on error
    
    Tasks can fork or die while we try to seize them. It's much more simpler
    and much more reliably to unseize what was seized and walk the tree again
    in case of some tree check failed.
    
    Yes, this makes it impossible to suspend a big tree which constantly forks :(
    but I'd prefer fixing more urgent issues before.
    
    Signed-off-by: Pavel Emelyanov <xemul at parallels.com>
    Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
---
 cr-dump.c |   37 ++++++++++++++++++++++++++++++++++++-
 1 files changed, 36 insertions(+), 1 deletions(-)

diff --git a/cr-dump.c b/cr-dump.c
index 0d5bb3a..36c5683 100644
--- a/cr-dump.c
+++ b/cr-dump.c
@@ -1012,7 +1012,42 @@ static int collect_subtree(pid_t pid, pid_t ppid, struct list_head *pstree_list,
 
 static int collect_pstree(pid_t pid, struct list_head *pstree_list, int leader_only)
 {
-	return collect_subtree(pid, -1, pstree_list, leader_only);
+	int ret, attempts = 5;
+
+	while (1) {
+		ret = collect_subtree(pid, -1, pstree_list, leader_only);
+		if (ret == 0)
+			break;
+
+		/*
+		 * Old tasks can die and new ones can appear while we
+		 * try to seize the swarm. It's much simpler (and reliable)
+		 * just to restart the collection from the beginning
+		 * rather than trying to chase them.
+		 */
+
+		if (attempts == 0)
+			break;
+
+		attempts--;
+		pr_info("Trying to suspend tasks again\n");
+
+		while (!list_empty(pstree_list)) {
+			struct pstree_item *item;
+
+			item = list_first_entry(pstree_list,
+					struct pstree_item, list);
+			list_del(&item->list);
+
+			unseize_task_and_threads(item, TASK_ALIVE);
+
+			xfree(item->children);
+			xfree(item->threads);
+			xfree(item);
+		}
+	}
+
+	return ret;
 }
 
 static int dump_pstree(pid_t pid, struct list_head *pstree_list, struct cr_fdset *cr_fdset)