fix CSNode array overflow issue

[NKTelnet]

If a node has not been insert into CSNode array before, when update() is called, we need to call insert() to check whether the CSNode array is full or not.

[ethouris]

1. What exactly is the predicted scenario that this function is called without inserting the node into the SndUList first?

2. Why do you make the first (added) insertion under the critical section, but the second one is outside the critical section?

[NKTelnet]

1. What exactly is the predicted scenario that this function is called without inserting the node into the SndUList first?

there is no SndUList::insert() in whole srtcore code before. Only SndUList::update() will be used.

2. Why do you make the first (added) insertion under the critical section, but the second one is outside the critical section?

because SndUList::insert() will use the same lock. the lock may be not reentry

[ethouris]

I saw that inserting is called somewhere in the sending function (in core.cpp). I still don't get what exactly problem this PR solves.

[NKTelnet]

I saw that inserting is called somewhere in the sending function (in core.cpp). I still don't get what exactly problem this PR solves.

Could you please show me the code ? I can not find it in master branch:

[root@ srtcore]# grep -r 'insert(' core.cpp
num = m_pSndLossList->insert(losslist_lo, losslist_hi);
num = m_pSndLossList->insert(m_iSndLastAck, losslist_hi);
int num = m_pSndLossList->insert(losslist[i], losslist[i]);
m_pRcvLossList->insert(seqlo, seqhi);
m_FreshLoss.insert(m_FreshLoss.begin() + i + 1, CRcvFreshLoss(next_begin, next_end, m_FreshLoss[i].ttl));
int num = m_pSndLossList->insert(m_iSndLastAck, csn);
int num = m_pSndLossList->insert(m_iSndLastAck, csn);
m_sPollID.insert(eid);
remove.insert(eid);

[ethouris]

Ok, it was something else. Looxlike "update" is the only function used for it, and "insert" is not in use at all. I must research it myself first.

[NKTelnet]

Ok, it was something else. Looxlike "update" is the only function used for it, and "insert" is not in use at all. I must research it myself first.

1. insert() has not been called.
2. the update() is wrong no matter whether insert() has been called or not.

[ethouris]

Ok, now I understand what's going on here.

Your change is perfectly ok, but I'd improve it with a small refactoring:

1. Change insert into checkEnlarge, remove the mutex lock and a call to insert_. Parameters will be empty in this case.
2. Simply call it before insert_ in the place where you wanted to call insert and forceUnlock earlier.
3. After that you can also simplify the structure by using else.

This will at least avoid the unnecessary unlock-lock cycle.