mirror of
https://github.com/torvalds/linux.git
synced 2026-11-20 13:03:48 +00:00
The "sector_size:<bytes>" option is capped at 4096 bytes. The reason is stated in commit8f0009a225("dm crypt: optionally support larger encryption sector size"): "the maximal IO must fit into the page limit, so the limit is set to the minimal page size possible (4096 bytes)." The rule is right; only the way it is resolved is not. The page limit it refers to is a property of the kernel that is running, but it was written as the smallest page size of any architecture, so a kernel with larger pages is held to a limit that belongs to a different one. The block layer has since stopped doing that for its own block size, in commit47dd675323("block/bdev: lift block size restrictions to 64k"), and dm-crypt is now the stricter of the two. Apply the same rule to the kernel being built: raise the cap to min(PAGE_SIZE, BLK_MAX_BLOCK_SIZE). PAGE_SIZE is still the page limit the original commit meant. A sector is passed to the crypto API as a single scatterlist entry, bio_iter_iovec() never returns more than PAGE_SIZE bytes, and crypt_alloc_buffer() may fall back to order-0 pages for the write bounce buffer. BLK_MAX_BLOCK_SIZE caps the logical block size that crypt_io_hints() announces. It is PAGE_SIZE without transparent hugepages, and 64K with them, which no architecture having a PAGE_SIZE above 64K can enable - so it does not lower the bound on any configuration today. It is in the expression so that this target cannot announce a block size blk_validate_limits() would reject. A larger unit also turns several crypto requests per page into one. Whether that is worth anything depends on the driver: for a CPU cipher the per-request cost is small, and it only pays off where a request carries a large fixed cost, as when the cipher is offloaded over DMA. Widen sector_size to unsigned int so that it can hold a sector larger than 65535. That also makes the option reject an argument of 69632, which %hu truncates to 4096 and accepts as a 4096-byte sector. Document that an encryption sector larger than the unit the device writes atomically can be torn by a power failure, and what each cipher mode does when that happens. The maximum stays 4096 wherever PAGE_SIZE is 4096, the default stays 512 bytes, and every table using a size from 512 to 4096 behaves as it did. The one behavioural change is the truncation above: an argument that wrapped into range is now rejected rather than silently accepted. A mapping larger than 4096 bytes can only be activated where PAGE_SIZE allows, so it is not suitable for portable on-disk formats. Bump the target version so that userspace can detect the new limit. Assisted-by: LLM Signed-off-by: Itai Handler <[email protected]> Signed-off-by: Mikulas Patocka <[email protected]>