All posts in Development

In Part 1 of this series, I discussed an issue we were having in one of our SharePoint 2013 farms and how I determined the issue was occurring because of a set of event receivers acting on the library.  In this post, I will discuss the code being used and what the final result was determined to be.  Stick around, it’s not what you think.

To be terribly honest nothing jumped out at me while looking over the code.  The initial review of the code indicated the issue could be around where the event receiver was trying to determine if the user adding the file was a member of the site owner group.  The original code was:

private void ProcessItemIfAuthorized(SPItemEventProperties properties, string errorMessage, bool isItemAdding)
{
    bool isFile;
    bool isSiteOwner = false;

    //Get the list of groups user is a part of.  Find if a part of site Owners
    SPSecurityManager secMgr = new SPSecurityManager();            
    List<string> spGroupsInWeb = secMgr.GetSPGroupsInWeb(properties.Web);

    //Loop through each group in the web.  If the group is the owner group, check to see if user exists within.
    foreach(string spGroup in spGroupsInWeb)
    {
        if(spGroup.Contains("<NAME OF OWNER GROUP>"))
        {
            isSiteOwner = secMgr.CheckIfUserInSPGroup(properties.Web, spGroup, secMgr.ParseUserIDFromClaim(properties.UserLoginName));
        }
    }

This seems pretty straight forward, but when “CheckIfUserInSPGroup” is called things aren’t quite as kosher.

public bool CheckIfUserInSPGroup(SPWeb spWeb, string groupUserName, string userName)
{
    bool maxReached = false;
    bool existsInGroup = false;

    if (!existsInGroup)
    {
        SPSecurity.RunWithElevatedPrivileges(delegate()
        {
            using (SPSite spSite = new SPSite(spWeb.Site.ID))
            {
                using (SPWeb currentWeb = spSite.OpenWeb(spWeb.ID))
                {
                    foreach (SPPrincipalInfo accountInfo in SPUtility.GetPrincipalsInGroup(currentWeb, groupUserName, int.MaxValue - 1, out maxReached))
                    {
                        //check to see if the account we are looking at is a nested group.  If so call the function again to dig deeper.
                        if (accountInfo.PrincipalType == SPPrincipalType.SecurityGroup)
                        {
                            existsInGroup = this.CheckIfUserInSPGroup(spWeb, accountInfo.LoginName, userName);

                            if (existsInGroup)
                            {
                                break;
                            }
                        }
                        else
                        {
                            if (this.ParseUserIDFromClaim(accountInfo.LoginName) == userName)
                            {
                                existsInGroup = true;
                                break;
                            }
                        }
                    }
                }
            }
        });

Again, normally not a huge issue, except that best practices state that you shouldn’t instantiate SPSite, SPWeb or SPList objects within an Event Receiver.  The reason for this is it causes extra database calls (more information here: https://msdn.microsoft.com/en-us/library/office/ee724407(v=office.14).aspx).  I thought this could be the culprit, but wasn’t convinced.  If this was the issue, why does it work fine for years and then suddenly stop working?  The reason the code is instantiating the SPSite and SPWeb object is it is used elsewhere in the solution and could be called by users who do not have the required access.  The same goes for the event receiver.  If I do not have access to control security groups in the site, I get an UnauthorizedAccessException.

So I thought, why not just use that.  We can safely assume that if the UnauthorizedAccessException error is thrown, the user is not in the Owners group.  So I updated the code with a try\catch (why one wasn’t already being used I don’t know) and added some logic into the catch.  Not generally the best method, but when used for targeted exceptions I believe acceptable IMHO.

//Loop through each group in the web.  If the group is the owner group, check to see if user exists within.
foreach(string spGroup in spGroupsInWeb)
{
    if(spGroup.Contains("<NAME OF OWNER GROUP>"))
    {
        try
        {
            isSiteOwner = secMgr.CheckIfUserInSPGroupEvntRcvr(properties.Web, spGroup, secMgr.ParseUserIDFromClaim(properties.UserLoginName));                        
        }
        //If authorization exception is fired, then we can assume the user is NOT in the owners group
        catch(UnauthorizedAccessException exNotAuthorized)
        {
            SPDiagnosticsService.Local.WriteTrace
                (
                    0,
                    new SPDiagnosticsCategory("App Security", TraceSeverity.High, EventSeverity.ErrorCritical),
                    TraceSeverity.High,
                    exNotAuthorized.ToString(),
                    string.Empty
                );

            isSiteOwner = false;
        }
        catch(Exception ex)
        {
            SPDiagnosticsService.Local.WriteTrace
                (
                    0,
                    new SPDiagnosticsCategory("App Security", TraceSeverity.High, EventSeverity.ErrorCritical),
                    TraceSeverity.High,
                    ex.ToString(),
                    string.Empty
                );

            //have to assume they don't have necassary access so files aren't added or deleted upon error
            isSiteOwner = false;
        }
    }
}

I also created a new method for the event receiver to call.  I couldn’t modify the existing one, as it contained valid logic to handle users without access and was being used elsewhere.

        public bool CheckIfUserInSPGroupEvntRcvr(SPWeb spWeb, string groupUserName, string userName)
        {
            bool maxReached = false;
            bool existsInGroup = false;

            if (!existsInGroup)
            {
                foreach (SPPrincipalInfo accountInfo in SPUtility.GetPrincipalsInGroup(spWeb, groupUserName, int.MaxValue - 1, out maxReached))
                {
                    //check to see if the account we are looking at is a nested group.  If so call the function again to dig deeper.
                    if (accountInfo.PrincipalType == SPPrincipalType.SecurityGroup)
                    {
                        existsInGroup = this.CheckIfUserInSPGroupEvntRcvr(spWeb, accountInfo.LoginName, userName);

                        if (existsInGroup)
                        {
                            break;
                        }
                    }
                    else
                    {
                        if (this.ParseUserIDFromClaim(accountInfo.LoginName) == userName)
                        {
                            existsInGroup = true;
                            break;
                        }
                    }
                }
            }

            return existsInGroup;
        }

So I moved the code into Pre-Prod and tried it out.  No change.  Still hanging, throwing errors and crashing the app pool.

Next step was to install Visual Studio into Pre-Prod and attach to the IIS Worker process.  I followed the code until it got into the newly created CheckIfUserInSPGroupEvntRcvr method.  There it stayed.  It kept looping through the AD users and groups within the SharePoint group.  As it was looping I watched the worker process memory usage grow and grow until it finally crashed again.  This didn’t make any sense as there are NOT that many users in these groups.

The Cause of it All

I took a look at the ownership group for the site I was testing with.  Like most (not all) of our project sites, it contained an AD group that contains our project team.  Let’s call that group All-Project.  All-Project had about a dozen users within it, however, there was an anomaly.  It also contained the Owners group from another project site.  This was an oddity.  I took a look at the Owner group and it also contained the same All-Project group. There was the culprit.

As you can see in the code above, it is designed for nested groups, so if the code hits a group it digs down to see if the nested group contains the user.  Because this Owner group was added (in error I found out while trying to figure out why it was there) to the All-Projects group, the code would dig into All-Projects then to the Owners group, from there back into the All-Projects group and then back into the Owners group… see where I am going with this?  By adding that single group to the All-Projects group in error an infinite recursion loop was created in the code.

The Final Fix

So the final fix was not an environmental change or a code modification.  It was simply to remove the Owners group from the All-Projects group.  Once that was done, the original code functioned as designed.  If this becomes a regular occurrence I will have to update the code to handle such an event, but in this case, I didn’t.  The farm is in containment (no further development short of break\fix) and the issue has not occurred for two years before this.  I hope the steps I documented in this blog series helps others out.

Thanks for reading!