38f691048 simplewallet: plug a timing leak (moneromooo-monero)
dcff02e4c epee: allow a random component in once_a_time timeouts (moneromooo-monero)
e10833024 wallet: reuse cached height when set after refresh (moneromooo-monero)
5956beaa1 wallet2: fix is_synced checking target height, not height (moneromooo-monero)
fd35e2304 wallet: fix another facet of "did I get some monero" information leak (moneromooo-monero)
d5472bd87 wallet2: do not send an unnecessary last getblocks.bin call on refresh (moneromooo-monero)
97ae7bb5c wallet2: do not repeatedly ask for pool txes sent to us (moneromooo-monero)
As reported by Tramèr et al, timing of refresh requests can be used
to see whether a password was requested (and thus at least one output
received) since this will induce a delay in subsequent calls.
To avoid this, we schedule calls at a given time instead of sleeping
for a set time (which would make delays additive).
To further avoid a scheduled call being during the time in which a
password is prompted, the actual scheduled time is now randomized.
Refreshing sets cached height, which is otherwise got by calling
get_info. Since get_info is called upon needing to display a prompt
after a command has finished, it can be used to determine how much
time a given command took to run if the cache timeout lapses while
the command runs. Refreshing caches the height as a side effect, so
get_info will never be called as a result of displaying a prompt
after refreshing (and potentially leaking how much time it took to
process a set of transactions, therefore leaking whether we got
some monero in them).
Target height would be appropriate for the daemon, which syncs
off other daemons, but the wallet syncs off the daemon it's
connected to, and its target is the daemon's current height.
We get new pool txes before processing any tx, pool or not.
This ensures that if we're asked for a password, this does not
cause a measurable delay in the txpool query after the last
block query.
The "everything refreshed" state was detected when a refresh call did
not return any new blocks. This can be detected without that extra
"empty" call by comparing the claimed node height to the height of
the last block retrieved. Doing this avoids that last call, saves
some bandwidth, and makes the common refresh case use only one call
rather than two.
As a side effect, it prevents an information leak reported by
Tramèr et al: if the wallet retrieves a set of blocks which includes
an output sent to the refreshing wallet, the wallet will prompt the
user for the password to decode the amount and calculate the key
image for the new output, and this will delay subsequent calls to
getblocks.bin, allowing a passive adversary to note the delay and
deduce when the wallet receives at least one output.
This can still happen if the wallet downloads more than 1000 blocks,
since this will be split in several calls, but then the most the
adversary can tell is which 1000 block section the user received
some monero (the adversary can estimate the heights of the blocks
by calculating how many "large" transfers are done, which will be
sections of blocks, the last of which will usually be below 1000,
but the size of the data should allow the actual number of blocks
sent to be determined fairly accurately).
This timing trick still be used via the subsequent scan for incoming
txes in the txpool, which will be fixed later.
This lets a passive attacker with access to the network link
between node and wallet perform traffic analysis to deduce
when an idle wallet receives a transaction.
Reported by Tramèr et al.
This allows flushing internal caches (for now, the bad tx cache,
which will allow debugging a stuck monerod after it has failed to
verify a transaction in a block, since it would otherwise not try
again, making subsequent log changes pointless)