On 26.10.2023 23:53, Alexei Starovoitov wrote:
On Wed, Oct 25, 2023 at 6:59 PM Vadim Fedorenko <vadfed@xxxxxxxx> wrote:+__bpf_kfunc struct bpf_crypto_skcipher_ctx * +bpf_crypto_skcipher_ctx_create(const struct bpf_dynptr_kern *algo, const struct bpf_dynptr_kern *key, + int *err) +{ + struct bpf_crypto_skcipher_ctx *ctx; + + if (__bpf_dynptr_size(algo) > CRYPTO_MAX_ALG_NAME) { + *err = -EINVAL; + return NULL; + } + + if (!crypto_has_skcipher(algo->data, CRYPTO_ALG_TYPE_SKCIPHER, CRYPTO_ALG_TYPE_MASK)) { + *err = -EOPNOTSUPP; + return NULL; + } + + ctx = bpf_mem_cache_alloc(&bpf_crypto_ctx_ma);Since this kfunc is sleepable, just kmalloc(GFP_KERNEL) here. No need to use bpf_mem_alloc.
I was thinking about adding GFP_ATOMIC allocation option to crypto_alloc_sync_skcipher, it's already implemented for cloning skcipher object. Then the code can be reused for both sleepable (expect module loading) and non-sleepable (fail if there is no crypto module loaded) variants without any changes. But I can implement different allocators for different options.
