Potential overflow in crowdsale participation with indivisible tokens

[dexX7]

CMPTransaction::logicHelper_CrowdsaleParticipation:

if (!isPropertyDivisible(property)) { // <-- the tokens "invested"
    nValue = nValue * 1e8; // <-- nValue is uint64_t
}

The cut-off is 92233720368, which is the largest number where nValue * 100000000 <= INT64_MAX, and sending more than 92233720368 indivisible tokens either causes an assertion violation or undefined behavior.

To be more specific:

Amount	Master Core 0.0.9	Omni Core 0.0.10
0 to 92233720368	OK	OK
92233720369 to 184467440737	OK	assertion violation
184467440738 or more	undefined	undefined

The assertion violation is caused by the conversion of uint64_t to int64_t, which results in negative numbers in that range (rejected via ConvertTo256).

The results for numbers higher than 184467440738 are undefined, meaning: it's up to the compiler implementation how it's dealt with. On my system it seems to be 0, but it can really be anything between 0 and MAX.

The positive:

• no new tokens can be created out of the blue
• we can easily mirror the exact behavior of 0.0.9

The negative:

• a buyer may not get the tokens he or she paid for

It's notable, even without overflow, that 92233720368 is the maximum number of divisible tokens that are "counted" anyway..

Potential fix:

int64_t adjusted = 0;

if (nValue > (INT64_MAX/100000000)) { // > 92233720368
    adjusted = INT64_MAX; // clip to max. allowed amount
} else {
    adjusted = nValue * 100000000;
}

edit: calculateFractional (for missed tokens) must be adjusted for the fix.

This is potentially consensus breaking.

[zathras-crypto]

Nice, thanks for this dude :)

Why don't we do the inflation of amounts once we're in uint256? We can avoid the modification of nvalue and overflow for INT64 should no longer be a concern?

With a simple patch such as zathras-crypto@33f3114, perhaps we may address this issue, #248 and #249 together?

Note - this is just calculateFundraiser, I haven't looked at calculateFractional yet...

I attempted to replicate your regtest with the patch:

omni_gettransaction af270a05abccd89c4ed7e554cf09c94da3cccad8f35851748f052f0b24e6844b
{
"txid" : "af270a05abccd89c4ed7e554cf09c94da3cccad8f35851748f052f0b24e6844b",
"fee" : "0.00000256",
"sendingaddress" : "n144os5tJMmfVP8SvRJD5TSZph7HMd7P7J",
"referenceaddress" : "msDqmnX8jpTRJdBevjqN5UQqfXUV9WU5aK",
"ismine" : true,
"version" : 0,
"type_int" : 0,
"type" : "Crowdsale Purchase",
"propertyid" : 3,
"divisible" : false,
"amount" : "1",
"purchasedpropertyid" : 4,
"purchasedpropertyname" : "Z_TestProperty",
"purchasedpropertydivisible" : false,
"purchasedtokens" : "1",
"issuertokens" : "0",
"valid" : true,
"blockhash" : "026a81dfeb9223dbf5ee5168556b60f01aa9066ee4820fbe4ee0dcf75d48252f",
"blocktime" : 1442730900,
"block" : 165,
"confirmations" : 1
}

omni_getcrowdsale 4 true
{
"propertyid" : 4,
"name" : "Z_TestProperty",
"active" : true,
"issuer" : "msDqmnX8jpTRJdBevjqN5UQqfXUV9WU5aK",
"propertyiddesired" : 3,
"tokensperunit" : "1",
"earlybonus" : 0,
"percenttoissuer" : 0,
"starttime" : 1442730900,
"deadline" : 1542729212,
"tokensissued" : "1",
"addedissuertokens" : "0",
"amountraised" : "1",
"participanttransactions" : [
{
"txid" : "af270a05abccd89c4ed7e554cf09c94da3cccad8f35851748f052f0b24e6844b",
"amountsent" : "1",
"participanttokens" : "1",
"issuertokens" : "0"
}
]
}

[dexX7]

That's pretty nice actually, and it even removes the artificial limit. Can you think of any downsides?

Note - this is just calculateFundraiser, I haven't looked at calculateFractional yet...

I'm very sure calculateFractional should be overhauled, because it basically re-runs all crowdsale participations to calculate totals.. as double. :/

But there is good news: unless I overlook something, we may almost completely remove it, because we already sumed all values.

int64_t missedTokens = (crowdsale.getUserCreated() * [bonus]) - crowdsale.getIssuerCreated();

Then another point I was wondering about: the "missed tokens" are only calculated, if a crowdsale expires, but not in case the crowdsale is manually closed. But it should, I assume? edit: nevermind, the missing tokens are calculated for manually closed crowdsales, too.