BBMap can report those statistics, though reads longer than 6kbp will be broken into 6kbp segments. For this purpose, that shouldn't matter, though it's worth noting that the shorter the fragments, the lower the apparent error rate is.
mapPacBio.sh in=reads.fastq maxlen=6000 out=mapped.sam ref=reference.fasta
That will report the rates of mismatches, insertions, and deletions.
It's impossible to make an accurate statement about the distribution of read lengths, since these are dependent on the length of the input DNA and how careful you treat the library. With the most common library prep you have reads of 6-10kb, with a long tail up to ~200kb.
However, with an adapted protocol, very careful handling and old-school DNA extraction you can get reads of up to 970kb, as reported by Nick Loman and Josh Quick.