Recent advances in highly tunable radios, along with new spectrum policies have paved the way for design of mesh networks over spectrum that is non-contiguous, and has diverse propagation characteristics. Notable examples are multi-band WiMAX mesh, wireless mesh over digital TV white-spaces. The design of multihop wireless networks that can tune into any part of such large spectrum and adapt channel width presents certain unique challenges: (i) different communication neighbors of a node and different interference patterns in different bands, (ii) different spectral efficiencies in different bands and (iii) the interplay between different layer functions such as routing, scheduling as a result of this. In this paper, we solve the problem of frequency aware spectrum allocation to mesh nodes operating over diverse and fragmented spectrum. We show that, our solution has immediate application to jointly optimizing flow rate, routing, and scheduling. Our algorithms have provable performance guarantees and are amenable to distributed implementation. Simulations over a 25 node grid topology under proportional fair allocation show 1.8x-2.5x (depending on the number of radios per node) throughput gains compared to a frequency-unaware scheme.