kern: update initial process load logic to do per-segment mapping/decompression

This commit is contained in:
Michael Scire 2023-10-11 04:26:30 -07:00
parent c81d787dc3
commit 545ddaf92c
3 changed files with 169 additions and 72 deletions

View file

@ -133,7 +133,7 @@ namespace ams::kern {
}
Result MakeCreateProcessParameter(ams::svc::CreateProcessParameter *out, bool enable_aslr) const;
Result Load(KProcessAddress address, const ams::svc::CreateProcessParameter &params, KProcessAddress src) const;
void Load(const KPageGroup &pg, KVirtualAddress data) const;
Result SetMemoryPermissions(KProcessPageTable &page_table, const ams::svc::CreateProcessParameter &params) const;
};

View file

@ -156,40 +156,14 @@ namespace ams::kern {
KPageGroup *process_pg = std::addressof(pg);
ON_SCOPE_EXIT { process_pg->Close(); };
/* Get the temporary region. */
const auto &temp_region = KMemoryLayout::GetTempRegion();
MESOSPHERE_ABORT_UNLESS(temp_region.GetEndAddress() != 0);
/* Map the process's memory into the temporary region. */
KProcessAddress temp_address = Null<KProcessAddress>;
MESOSPHERE_R_ABORT_UNLESS(Kernel::GetKernelPageTable().MapPageGroup(std::addressof(temp_address), pg, temp_region.GetAddress(), temp_region.GetSize() / PageSize, KMemoryState_Kernel, KMemoryPermission_KernelReadWrite));
/* Setup the new page group's memory, so that we can load the process. */
{
/* Copy the unaligned ending of the compressed binary. */
if (const size_t unaligned_size = binary_size - util::AlignDown(binary_size, PageSize); unaligned_size != 0) {
std::memcpy(GetVoidPointer(temp_address + process_size - unaligned_size), GetVoidPointer(data + binary_size - unaligned_size), unaligned_size);
}
/* Copy the aligned part of the compressed binary. */
if (const size_t aligned_size = util::AlignDown(binary_size, PageSize); aligned_size != 0 && src_pool == dst_pool) {
std::memmove(GetVoidPointer(temp_address + process_size - binary_size), GetVoidPointer(temp_address), aligned_size);
} else {
if (src_pool != dst_pool) {
std::memcpy(GetVoidPointer(temp_address + process_size - binary_size), GetVoidPointer(data), aligned_size);
Kernel::GetMemoryManager().Close(KMemoryLayout::GetLinearPhysicalAddress(data), aligned_size / PageSize);
}
}
/* Clear the first part of the memory. */
std::memset(GetVoidPointer(temp_address), 0, process_size - binary_size);
}
/* Load the process. */
MESOSPHERE_R_ABORT_UNLESS(reader.Load(temp_address, params, temp_address + process_size - binary_size));
reader.Load(pg, data);
/* Unmap the temporary mapping. */
MESOSPHERE_R_ABORT_UNLESS(Kernel::GetKernelPageTable().UnmapPageGroup(temp_address, pg, KMemoryState_Kernel));
/* If necessary, close/release the aligned part of the data we just loaded. */
if (const size_t aligned_bin_size = util::AlignDown(binary_size, PageSize); aligned_bin_size != 0 && src_pool != dst_pool) {
Kernel::GetMemoryManager().Close(KMemoryLayout::GetLinearPhysicalAddress(data), aligned_bin_size / PageSize);
Kernel::GetSystemResourceLimit().Release(ams::svc::LimitableResource_PhysicalMemoryMax, aligned_bin_size);
}
/* Create a KProcess object. */
new_process = KProcess::Create();
@ -242,11 +216,6 @@ namespace ams::kern {
MESOSPHERE_R_ABORT_UNLESS(new_process->Initialize(params, *process_pg, reader.GetCapabilities(), reader.GetNumCapabilities(), std::addressof(Kernel::GetSystemResourceLimit()), dst_pool, reader.IsImmortal()));
}
/* Release the memory that was previously reserved. */
if (const size_t aligned_bin_size = util::AlignDown(binary_size, PageSize); aligned_bin_size != 0 && src_pool != dst_pool) {
Kernel::GetSystemResourceLimit().Release(ams::svc::LimitableResource_PhysicalMemoryMax, aligned_bin_size);
}
/* Set the process's memory permissions. */
MESOSPHERE_R_ABORT_UNLESS(reader.SetMemoryPermissions(new_process->GetPageTable(), params));

View file

@ -73,6 +73,119 @@ namespace ams::kern {
}
}
NOINLINE void LoadInitialProcessSegment(const KPageGroup &pg, size_t seg_offset, size_t seg_size, size_t binary_size, KVirtualAddress data, bool compressed) {
/* Save the original binary extents, for later use. */
const KPhysicalAddress binary_phys = KMemoryLayout::GetLinearPhysicalAddress(data);
/* Create a page group representing the segment. */
KPageGroup segment_pg(Kernel::GetSystemSystemResource().GetBlockInfoManagerPointer());
if (size_t remaining_size = util::AlignUp(seg_size, PageSize); remaining_size != 0) {
/* Find the pages whose data corresponds to the segment. */
size_t cur_offset = 0;
for (auto it = pg.begin(); it != pg.end() && remaining_size > 0; ++it) {
/* Get the current size. */
const size_t cur_size = it->GetSize();
/* Determine if the offset is in range. */
const size_t rel_diff = seg_offset - cur_offset;
const bool is_before = cur_offset <= seg_offset;
cur_offset += cur_size;
if (is_before && seg_offset < cur_offset) {
/* It is, so add the block. */
const size_t block_size = std::min<size_t>(cur_size - rel_diff, remaining_size);
MESOSPHERE_R_ABORT_UNLESS(segment_pg.AddBlock(it->GetAddress() + rel_diff, block_size / PageSize));
/* Advance. */
cur_offset = seg_offset + block_size;
remaining_size -= block_size;
seg_offset += block_size;
}
}
}
/* Setup the new page group's memory so that we can load the segment. */
{
KVirtualAddress last_block = Null<KVirtualAddress>;
KVirtualAddress last_data = Null<KVirtualAddress>;
size_t last_copy_size = 0;
size_t last_clear_size = 0;
size_t remaining_copy_size = binary_size;
for (const auto &block : segment_pg) {
/* Get the current block extents. */
const auto block_addr = block.GetAddress();
const size_t block_size = block.GetSize();
if (remaining_copy_size > 0) {
/* Determine if we need to copy anything. */
const size_t cur_size = std::min<size_t>(block_size, remaining_copy_size);
/* NOTE: The first block may potentially overlap the binary we want to copy to. */
/* Consider e.g. the case where the overall compressed image has size 0x40000, seg_offset is 0x30000, and binary_size is > 0x20000. */
/* Suppose too that data points, say, 0x18000 into the compressed image. */
/* Suppose finally that we simply naively copy in order. */
/* The first iteration of this loop will perform an 0x10000 copy from image+0x18000 to image + 0x30000 (as there is no overlap). */
/* The second iteration will perform a copy from image+0x28000 to <allocated pages>. */
/* However, the first copy will have trashed the data in the second copy. */
/* Thus, we must copy the first block after-the-fact to avoid potentially trashing data in the overlap case. */
/* It is guaranteed by pre-condition that only the very first block can overlap with the physical binary, so we can simply memmove it at the end. */
if (last_block != Null<KVirtualAddress>) {
/* This is guaranteed by pre-condition, but for ease of debugging, check for no overlap. */
AMS_ASSERT(!util::HasOverlap(GetInteger(binary_phys), binary_size, GetInteger(block_addr), cur_size));
/* We need to copy. */
std::memcpy(GetVoidPointer(KMemoryLayout::GetLinearVirtualAddress(block_addr)), GetVoidPointer(data), cur_size);
/* If we need to, clear past where we're copying. */
if (cur_size != block_size) {
std::memset(GetVoidPointer(KMemoryLayout::GetLinearVirtualAddress(block_addr + cur_size)), 0, block_size - cur_size);
}
/* Advance. */
remaining_copy_size -= cur_size;
data += cur_size;
} else {
/* Save the first block, which may potentially overlap, so that we can copy it later. */
last_block = KMemoryLayout::GetLinearVirtualAddress(block_addr);
last_data = data;
last_copy_size = cur_size;
last_clear_size = block_size - cur_size;
/* Advance. */
remaining_copy_size -= cur_size;
data += cur_size;
}
} else {
/* We don't have data to copy, so we should just clear the pages. */
std::memset(GetVoidPointer(KMemoryLayout::GetLinearVirtualAddress(block_addr)), 0, block_size);
}
}
/* Handle a last block. */
if (last_copy_size != 0) {
if (last_block != last_data) {
std::memmove(GetVoidPointer(last_block), GetVoidPointer(last_data), last_copy_size);
}
if (last_clear_size != 0) {
std::memset(GetVoidPointer(last_block + last_copy_size), 0, last_clear_size);
}
}
}
/* If compressed, uncompress the data. */
if (compressed) {
/* Get the temporary region. */
const auto &temp_region = KMemoryLayout::GetTempRegion();
MESOSPHERE_ABORT_UNLESS(temp_region.GetEndAddress() != 0);
/* Map the process's memory into the temporary region. */
KProcessAddress temp_address = Null<KProcessAddress>;
MESOSPHERE_R_ABORT_UNLESS(Kernel::GetKernelPageTable().MapPageGroup(std::addressof(temp_address), segment_pg, temp_region.GetAddress(), temp_region.GetSize() / PageSize, KMemoryState_Kernel, KMemoryPermission_KernelReadWrite));
ON_SCOPE_EXIT { MESOSPHERE_R_ABORT_UNLESS(Kernel::GetKernelPageTable().UnmapPageGroup(temp_address, segment_pg, KMemoryState_Kernel)); };
/* Uncompress the data. */
BlzUncompress(GetVoidPointer(temp_address + binary_size));
}
}
}
Result KInitialProcessReader::MakeCreateProcessParameter(ams::svc::CreateProcessParameter *out, bool enable_aslr) const {
@ -113,11 +226,13 @@ namespace ams::kern {
MESOSPHERE_ABORT_UNLESS(start_address == 0);
/* Set fields in parameter. */
out->code_address = map_start + start_address;
out->code_num_pages = util::AlignUp(end_address - start_address, PageSize) / PageSize;
out->program_id = m_kip_header.GetProgramId();
out->version = m_kip_header.GetVersion();
out->flags = 0;
out->code_address = map_start + start_address;
out->code_num_pages = util::AlignUp(end_address - start_address, PageSize) / PageSize;
out->program_id = m_kip_header.GetProgramId();
out->version = m_kip_header.GetVersion();
out->flags = 0;
out->reslimit = ams::svc::InvalidHandle;
out->system_resource_num_pages = 0;
MESOSPHERE_ABORT_UNLESS((out->code_address / PageSize) + out->code_num_pages <= (map_end / PageSize));
/* Copy name field. */
@ -146,42 +261,55 @@ namespace ams::kern {
R_SUCCEED();
}
Result KInitialProcessReader::Load(KProcessAddress address, const ams::svc::CreateProcessParameter &params, KProcessAddress src) const {
void KInitialProcessReader::Load(const KPageGroup &pg, KVirtualAddress data) const {
/* Prepare to layout the data. */
const KProcessAddress rx_address = address + m_kip_header.GetRxAddress();
const KProcessAddress ro_address = address + m_kip_header.GetRoAddress();
const KProcessAddress rw_address = address + m_kip_header.GetRwAddress();
const u8 *rx_binary = GetPointer<const u8>(src);
const u8 *ro_binary = rx_binary + m_kip_header.GetRxCompressedSize();
const u8 *rw_binary = ro_binary + m_kip_header.GetRoCompressedSize();
const KVirtualAddress rx_data = data;
const KVirtualAddress ro_data = rx_data + m_kip_header.GetRxCompressedSize();
const KVirtualAddress rw_data = ro_data + m_kip_header.GetRoCompressedSize();
const size_t rx_size = m_kip_header.GetRxSize();
const size_t ro_size = m_kip_header.GetRoSize();
const size_t rw_size = m_kip_header.GetRwSize();
/* Copy text. */
if (util::AlignUp(m_kip_header.GetRxSize(), PageSize)) {
std::memmove(GetVoidPointer(rx_address), rx_binary, m_kip_header.GetRxCompressedSize());
if (m_kip_header.IsRxCompressed()) {
BlzUncompress(GetVoidPointer(rx_address + m_kip_header.GetRxCompressedSize()));
/* If necessary, setup bss. */
if (const size_t bss_size = m_kip_header.GetBssSize(); bss_size > 0) {
/* Determine how many additional pages are needed for bss. */
const u64 rw_end = util::AlignUp<u64>(m_kip_header.GetRwAddress() + m_kip_header.GetRwSize(), PageSize);
const u64 bss_end = util::AlignUp<u64>(m_kip_header.GetBssAddress() + m_kip_header.GetBssSize(), PageSize);
if (rw_end != bss_end) {
/* Find the pages corresponding to bss. */
size_t cur_offset = 0;
size_t remaining_size = bss_end - rw_end;
size_t bss_offset = rw_end - m_kip_header.GetRxAddress();
for (auto it = pg.begin(); it != pg.end() && remaining_size > 0; ++it) {
/* Get the current size. */
const size_t cur_size = it->GetSize();
/* Determine if the offset is in range. */
const size_t rel_diff = bss_offset - cur_offset;
const bool is_before = cur_offset <= bss_offset;
cur_offset += cur_size;
if (is_before && bss_offset < cur_offset) {
/* It is, so clear the bss range. */
const size_t block_size = std::min<size_t>(cur_size - rel_diff, remaining_size);
std::memset(GetVoidPointer(KMemoryLayout::GetLinearVirtualAddress(it->GetAddress() + rel_diff)), 0, block_size);
/* Advance. */
cur_offset = bss_offset + block_size;
remaining_size -= block_size;
bss_offset += block_size;
}
}
}
}
/* Copy rodata. */
if (util::AlignUp(m_kip_header.GetRoSize(), PageSize)) {
std::memmove(GetVoidPointer(ro_address), ro_binary, m_kip_header.GetRoCompressedSize());
if (m_kip_header.IsRoCompressed()) {
BlzUncompress(GetVoidPointer(ro_address + m_kip_header.GetRoCompressedSize()));
}
}
/* Load .rwdata. */
LoadInitialProcessSegment(pg, m_kip_header.GetRwAddress() - m_kip_header.GetRxAddress(), rw_size, m_kip_header.GetRwCompressedSize(), rw_data, m_kip_header.IsRwCompressed());
/* Copy rwdata. */
if (util::AlignUp(m_kip_header.GetRwSize(), PageSize)) {
std::memmove(GetVoidPointer(rw_address), rw_binary, m_kip_header.GetRwCompressedSize());
if (m_kip_header.IsRwCompressed()) {
BlzUncompress(GetVoidPointer(rw_address + m_kip_header.GetRwCompressedSize()));
}
}
/* Load .rodata. */
LoadInitialProcessSegment(pg, m_kip_header.GetRoAddress() - m_kip_header.GetRxAddress(), ro_size, m_kip_header.GetRoCompressedSize(), ro_data, m_kip_header.IsRoCompressed());
MESOSPHERE_UNUSED(params);
R_SUCCEED();
/* Load .text. */
LoadInitialProcessSegment(pg, m_kip_header.GetRxAddress() - m_kip_header.GetRxAddress(), rx_size, m_kip_header.GetRxCompressedSize(), rx_data, m_kip_header.IsRxCompressed());
}
Result KInitialProcessReader::SetMemoryPermissions(KProcessPageTable &page_table, const ams::svc::CreateProcessParameter &params) const {