In his book "Foundations of Computing", professor Allison shows an example of "language of equal numbers of as, bs, and cs, but in any order", formally: $L = \{ w \in \{a,b,c\}^*\ |\ |w|_a=|w|_b=|w|_c \}$, where $|w|_a$ denotes the number of $a$s in a word $w$. This language is a variant of the language $L_1 = \{ a^nb^nc^n |\ n>0 \}$ which is context-sensitive.
Later, he says that this language is not context-sensitive because "the working strings grow beyond a constant times the size of the initial input before shrinking down to the final output."
However, I can think of a Turing machine that first sorts the input (in a linear space n + c, where n is the input size and c is a constant such as 1) and then accepts/rejects in the very same manner as the machine accepting the language $L_1$. Both subroutines are also linear bounded and thus the machine itself is a linear bounded automaton (LBA). Context-sensitive languages are accepted by LBAs, therefore the language $L$ is context-sensitive.
Where am I mistaken?