Kernel/FATFS: Only read the requested blocks in read_bytes_locked()

This dramatically improves performance when working with large files,
since we no longer re-read the entire file for each read.
This commit is contained in:
implicitfield 2024-04-21 01:11:55 +04:00 committed by Tim Schumacher
parent 5bc87ad1a5
commit c15b473c1a

View file

@ -434,19 +434,29 @@ ErrorOr<size_t> FATInode::read_bytes_locked(off_t offset, size_t size, UserOrKer
if (offset >= m_entry.file_size)
return 0;
// FIXME: Read only the needed blocks instead of the whole file
auto blocks = TRY(const_cast<FATInode&>(*this).read_block_list());
auto block_list = TRY(const_cast<FATInode&>(*this).get_block_list());
// Take the minimum of the:
// 1. User-specified size parameter
// 2. The file size.
// 3. The number of blocks returned for reading.
size_t read_size = min(
min(size, m_entry.file_size - offset),
(m_cluster_list.size() * fs().m_device_block_size * fs().m_parameter_block->common_bpb()->sectors_per_cluster) - offset);
TRY(buffer.write(blocks->data() + offset, read_size));
u32 first_block_index = offset / fs().m_device_block_size;
u32 last_block_index = (offset + size - 1) / fs().m_device_block_size;
return read_size;
size_t offset_into_first_block = offset - first_block_index * fs().m_device_block_size;
size_t nread = 0;
size_t remaining_count = size;
for (u32 block_index = first_block_index; block_index <= last_block_index; ++block_index) {
size_t offset_into_block = block_index == first_block_index ? offset_into_first_block : 0;
size_t to_read = min(fs().m_device_block_size - offset_into_block, remaining_count);
auto buffer_offset = buffer.offset(nread);
dbgln_if(FAT_DEBUG, "FATInode[{}]::read_bytes_locked(): Reading {} byte(s) from block {} at offset {}", identifier(), to_read, block_list[block_index], offset_into_block);
TRY(fs().read_block(block_list[block_index], &buffer_offset, to_read, offset_into_block));
nread += to_read;
remaining_count -= to_read;
}
return size;
}
InodeMetadata FATInode::metadata() const