For contaminant filtering, I find it is beneficial to remove both reads in a pair if either maps to the contaminant genome. For that you can use BBMap like this (assuming the reads are interleaved in a single file; if not you can use "in1" and "in2"):

bbmap.sh in=reads.fq ref=contam.fasta outm=bad.sam outu=good.fq

Here, "good" will have all the paired reads in which neither mapped to the reference, and they will still be properly interleaved with their original names, which will not be the case if you use samtools to do the filtering.

First align your reads and get your aligned reads

samtools view -b -F4 in.bam > out_mapped.bam

then convert your bam in fastq (with bamtools for example)

and re-align with your list of contaminant and remove aligned reads

samtools view -b -f4 in.bam > out_unmapped.bam