On Wed, 5 Jul 2023, Eric Biggers wrote: > On Wed, Jul 05, 2023 at 05:40:08PM +0100, Giovanni Cabiddu wrote: > > > Algorithms that do not set this flag will guarantee > > "will guarantee" => "guarantee" > > > that memory is not allocated during request processing, except in > > the avoidable exception cases described below. > > "avoidable exception cases" => "exception cases" > > Whether they are avoidable depends on the user. > > > * Users can request an algorithm with this flag unset if they can't handle > > * memory allocation failures or sleeping during request processing. > > Why add the "sleeping during request processing" part? Isn't that controlled on > a per-request basis by CRYPTO_TFM_REQ_MAY_SLEEP which is a separate thing? > > > * They should also follow the constraints below. > > "should" => "must" > > > + * - The input and output scatterlists must have no more than 4 entries. > > + * If the scatterlists contain more than 4 entries, the algorithm may > > + * allocate memory. > > "If the scatterlists contains" => "If either scatterlist contains" > > Otherwise it is unclear whether this is talking about the length of each > scatterlist individually, or the sum of their lengths. > > - Eric Hi I wouldn't change the meaning of CRYPTO_ALG_ALLOCATES_MEMORY (because people will forget about this subtle change anyway). Also note that dm-integrity allocates arbitrarily large sg-lists when encrypting the journal, so if you change the meaning of CRYPTO_ALG_ALLOCATES_MEMORY, there would be no flag left for dm-integrity to test. I would introduce a new flag, something like CRYPTO_ALG_ALLOCATES_MEMORY_FOR_5_OR_MORE_SG_ENTRIES. dm-crypt can then filter the algorithms based on this flag - and the rest of the kernel code may stay unchanged. Mikulas
