Kernel/Userspace: Replace my custom SHM system with SysV shm

This allows xbanan to use shm!
This commit is contained in:
2026-08-08 01:52:12 +03:00
parent 2c4fee4a82
commit fc40820150
23 changed files with 327 additions and 225 deletions
+1 -1
View File
@@ -59,7 +59,7 @@ signal_trampoline:
addl $24, %esp
// restore sigmask
movl $78, %eax // SYS_SIGPROCMASK
movl $79, %eax // SYS_SIGPROCMASK
movl $3, %ebx // SIG_SETMASK
leal 72(%esp), %ecx // set
xorl %edx, %edx // oset
+1 -1
View File
@@ -61,7 +61,7 @@ signal_trampoline:
addq $40, %rsp
// restore sigmask
movq $78, %rdi // SYS_SIGPROCMASK
movq $79, %rdi // SYS_SIGPROCMASK
movq $3, %rsi // SIG_SETMASK
leaq 192(%rsp), %rdx // set
xorq %r10, %r10 // oset
@@ -6,25 +6,20 @@
#include <kernel/Lock/SpinLock.h>
#include <kernel/Memory/MemoryRegion.h>
#include <fcntl.h>
#include <sys/shm.h>
namespace Kernel
{
class SharedMemoryObject;
class SharedMemoryObjectManager
{
public:
using Key = size_t;
public:
static BAN::ErrorOr<void> initialize();
static SharedMemoryObjectManager& get();
BAN::ErrorOr<Key> create_object(size_t size, PageTable::flags_t);
BAN::ErrorOr<void> delete_object(Key);
BAN::ErrorOr<BAN::UniqPtr<SharedMemoryObject>> map_object(Key, PageTable&, AddressRange);
BAN::ErrorOr<int> shmget(key_t key, size_t size, int shmflg);
BAN::ErrorOr<void> shmctl(int shmid, int cmd, struct shmid_ds* user_buf);
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> shmat(int shmid, const void* shmaddr, int shmflg);
private:
SharedMemoryObjectManager() {}
@@ -32,49 +27,29 @@ namespace Kernel
private:
struct Object : public BAN::RefCounted<Object>
{
Object(key_t key, shmid_ds info)
: key(key)
, info(info)
{ }
~Object();
Key key;
size_t size;
PageTable::flags_t flags;
bool can_current_process_access(int flags) const;
const key_t key;
shmid_ds info;
Mutex mutex;
BAN::Vector<paddr_t> paddrs;
SpinLock spin_lock;
bool marked_for_deletion { false };
};
private:
Mutex m_mutex;
BAN::HashMap<Key, BAN::RefPtr<Object>> m_objects;
BAN::HashMap<key_t, int> m_ids;
BAN::HashMap<int, BAN::RefPtr<Object>> m_objects;
friend class SharedMemoryObject;
friend class BAN::UniqPtr<SharedMemoryObjectManager>;
};
class SharedMemoryObject : public MemoryRegion
{
BAN_NON_COPYABLE(SharedMemoryObject);
BAN_NON_MOVABLE(SharedMemoryObject);
public:
static BAN::ErrorOr<BAN::UniqPtr<SharedMemoryObject>> create(BAN::RefPtr<SharedMemoryObjectManager::Object>, PageTable&, AddressRange);
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> clone(PageTable& new_page_table) override;
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> split(size_t offset) override;
BAN::ErrorOr<void> msync(vaddr_t, size_t, int) override { return {}; }
protected:
BAN::ErrorOr<bool> allocate_page_containing_impl(vaddr_t vaddr, bool wants_write) override;
private:
SharedMemoryObject(BAN::RefPtr<SharedMemoryObjectManager::Object> object, PageTable& page_table)
: MemoryRegion(page_table, object->size, MemoryRegion::Type::SHARED, object->flags, O_EXEC | O_RDWR)
, m_object(object)
{ }
private:
BAN::RefPtr<SharedMemoryObjectManager::Object> m_object;
friend class BAN::UniqPtr<SharedMemoryObject>;
};
}
+6 -5
View File
@@ -179,9 +179,10 @@ namespace Kernel
BAN::ErrorOr<long> sys_mprotect(void* addr, size_t len, int prot);
BAN::ErrorOr<long> sys_msync(void* addr, size_t len, int flags);
BAN::ErrorOr<long> sys_smo_create(size_t len, int prot);
BAN::ErrorOr<long> sys_smo_delete(SharedMemoryObjectManager::Key);
BAN::ErrorOr<long> sys_smo_map(SharedMemoryObjectManager::Key);
BAN::ErrorOr<long> sys_shmget(key_t key, size_t size, int shmflg);
BAN::ErrorOr<long> sys_shmctl(int shmid, int cmd, struct shmid_ds* buf);
BAN::ErrorOr<long> sys_shmat(int shmid, const void* shmaddr, int shmflg);
BAN::ErrorOr<long> sys_shmdt(const void* shmaddr);
BAN::ErrorOr<long> sys_ttyname(int fildes, char* name, size_t namesize);
BAN::ErrorOr<long> sys_posix_openpt(int flags);
@@ -255,6 +256,8 @@ namespace Kernel
const VirtualFileSystem::File& working_directory() const { return m_working_directory; }
const VirtualFileSystem::File& root_file() const { return m_root_file; }
BAN::ErrorOr<AddressRange> find_free_address_range(size_t size);
private:
Process(const Credentials&, pid_t pid, pid_t parent, pid_t sid, pid_t pgrp);
static Process* create_process(const Credentials&, pid_t parent, pid_t sid = 0, pid_t pgrp = 0);
@@ -280,8 +283,6 @@ namespace Kernel
// You must hold reader end of m_mapped_region_lock when calling this.
size_t find_mapped_region(vaddr_t) const;
BAN::ErrorOr<AddressRange> find_free_address_range(size_t size);
BAN::ErrorOr<VirtualFileSystem::File> find_file(int fd, const char* path, int flags) const;
BAN::ErrorOr<FileParent> find_parent_file(int fd, const char* path, int flags) const;
BAN::ErrorOr<VirtualFileSystem::File> find_relative_parent(int fd, const char* path) const;
+186 -36
View File
@@ -1,13 +1,44 @@
#include <kernel/Lock/LockGuard.h>
#include <kernel/Memory/Heap.h>
#include <kernel/Memory/SharedMemoryObject.h>
#include <kernel/Process.h>
#include <kernel/Random.h>
#include <kernel/Timer/Timer.h>
#include <kernel/UserCopy.h>
#include <fcntl.h>
namespace Kernel
{
static BAN::UniqPtr<SharedMemoryObjectManager> s_instance;
class SharedMemoryObject : public MemoryRegion
{
BAN_NON_COPYABLE(SharedMemoryObject);
BAN_NON_MOVABLE(SharedMemoryObject);
public:
static BAN::ErrorOr<BAN::UniqPtr<SharedMemoryObject>> create(BAN::RefPtr<SharedMemoryObjectManager::Object>, PageTable&, AddressRange, int status_flags);
~SharedMemoryObject();
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> clone(PageTable& new_page_table) override;
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> split(size_t offset) override;
BAN::ErrorOr<void> msync(vaddr_t, size_t, int) override { return {}; }
protected:
BAN::ErrorOr<bool> allocate_page_containing_impl(vaddr_t vaddr, bool wants_write) override;
private:
SharedMemoryObject(BAN::RefPtr<SharedMemoryObjectManager::Object>, PageTable& page_table, int status_flags);
private:
BAN::RefPtr<SharedMemoryObjectManager::Object> m_object;
friend class BAN::UniqPtr<SharedMemoryObject>;
};
BAN::ErrorOr<void> SharedMemoryObjectManager::initialize()
{
ASSERT(!s_instance);
@@ -28,70 +59,189 @@ namespace Kernel
Heap::get().release_page(paddr);
}
BAN::ErrorOr<SharedMemoryObjectManager::Key> SharedMemoryObjectManager::create_object(size_t size, PageTable::flags_t flags)
BAN::ErrorOr<int> SharedMemoryObjectManager::shmget(key_t key, size_t size, int shmflg)
{
ASSERT(size % PAGE_SIZE == 0);
auto object = TRY(BAN::RefPtr<Object>::create());
object->size = size;
object->flags = flags;
TRY(object->paddrs.resize(size / PAGE_SIZE, 0));
LockGuard _(m_mutex);
// NOTE: don't set the top bit so cast to signed is not negative
auto generate_key = []() { return Random::get<Key>() & (~(Key)0 >> 1); };
if (key != IPC_PRIVATE)
{
if (auto it = m_ids.find(key); it != m_ids.end())
{
if (shmflg & IPC_EXCL)
return BAN::Error::from_errno(EEXIST);
Key key = generate_key();
while (m_objects.contains(key))
key = generate_key();
object->key = key;
const auto& object = m_objects[it->value];
if (object->info.shm_segsz < size)
return BAN::Error::from_errno(EINVAL);
TRY(m_objects.insert(key, object));
return key;
const int flags =
(shmflg & S_IRUSR ? O_RDONLY : 0) |
(shmflg & S_IWUSR ? O_WRONLY : 0);
if (!object->can_current_process_access(flags))
return BAN::Error::from_errno(EACCES);
return it->value;
}
if (!(shmflg & IPC_CREAT))
return BAN::Error::from_errno(ENOENT);
}
const auto& process = Process::current();
const uid_t uid = process.credentials().euid();
const gid_t gid = process.credentials().egid();
const pid_t pid = process.pid();
const mode_t mode = shmflg & 0777;
auto object = TRY(BAN::RefPtr<Object>::create(key, shmid_ds {
.shm_perm = {
.uid = uid,
.gid = gid,
.cuid = uid,
.cgid = uid,
.mode = mode,
},
.shm_segsz = size,
.shm_lpid = 0,
.shm_cpid = pid,
.shm_nattch = 0,
.shm_atime = 0,
.shm_dtime = 0,
.shm_ctime = SystemTimer::get().real_time().tv_sec,
}));
TRY(object->paddrs.resize(BAN::Math::div_round_up(size, PAGE_SIZE), 0));
auto generate_id = []() { return Random::get<unsigned>() & BAN::numeric_limits<int>::max(); };
int shmid = generate_id();
while (m_ids.contains(shmid))
shmid = generate_id();
if (key != IPC_PRIVATE)
TRY(m_ids.insert(key, shmid));
if (auto ret = m_objects.insert(shmid, object); ret.is_error())
{
if (key != IPC_PRIVATE)
m_ids.remove(key);
return ret.release_error();
}
return shmid;
}
BAN::ErrorOr<void> SharedMemoryObjectManager::delete_object(Key key)
BAN::ErrorOr<void> SharedMemoryObjectManager::shmctl(int shmid, int cmd, struct shmid_ds* user_buf)
{
LockGuard _(m_mutex);
LockGuard _0(m_mutex);
auto it = m_objects.find(key);
auto it = m_objects.find(shmid);
if (it == m_objects.end())
return BAN::Error::from_errno(ENOENT);
return BAN::Error::from_errno(EINVAL);
LockGuard _1(it->value->mutex);
switch (cmd)
{
case IPC_RMID:
if (!it->value->can_current_process_access(O_WRONLY))
return BAN::Error::from_errno(EACCES);
if (it->key != IPC_PRIVATE)
m_ids.remove(it->key);
if (it->value->info.shm_nattch == 0)
m_objects.remove(it);
else
it->value->marked_for_deletion = true;
break;
case IPC_STAT:
TRY(write_to_user(user_buf, &it->value->info, sizeof(shmid_ds)));
break;
case IPC_SET:
{
if (!it->value->can_current_process_access(O_WRONLY))
return BAN::Error::from_errno(EACCES);
shmid_ds buf;
TRY(read_from_user(user_buf, &buf, sizeof(shmid_ds)));
it->value->info.shm_perm.uid = buf.shm_perm.uid;
it->value->info.shm_perm.gid = buf.shm_perm.gid;
it->value->info.shm_perm.mode = buf.shm_perm.mode & 0777;
break;
}
default:
return BAN::Error::from_errno(EINVAL);
}
m_objects.remove(it);
return {};
}
BAN::ErrorOr<BAN::UniqPtr<SharedMemoryObject>> SharedMemoryObjectManager::map_object(Key key, PageTable& page_table, AddressRange address_range)
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> SharedMemoryObjectManager::shmat(int shmid, const void* shmaddr, int shmflg)
{
LockGuard _(m_mutex);
auto it = m_objects.find(key);
auto it = m_objects.find(shmid);
if (it == m_objects.end())
return BAN::Error::from_errno(ENOENT);
return TRY(SharedMemoryObject::create(it->value, page_table, address_range));
const int status_flags = (shmflg & SHM_RDONLY) ? O_RDONLY : O_RDWR;
if (!it->value->can_current_process_access(status_flags))
return BAN::Error::from_errno(EACCES);
const AddressRange address_range = (shmaddr == nullptr)
? AddressRange { .start = 0x400000, .end = USERSPACE_END }
: TRY(Process::current().find_free_address_range(it->value->info.shm_segsz));
auto region = TRY(SharedMemoryObject::create(it->value, Process::current().page_table(), address_range, status_flags));
return BAN::UniqPtr<MemoryRegion>(BAN::move(region));
}
BAN::ErrorOr<BAN::UniqPtr<SharedMemoryObject>> SharedMemoryObject::create(BAN::RefPtr<SharedMemoryObjectManager::Object> object, PageTable& page_table, AddressRange address_range)
bool SharedMemoryObjectManager::Object::can_current_process_access(int flags) const
{
auto smo = TRY(BAN::UniqPtr<SharedMemoryObject>::create(object, page_table));
TRY(smo->initialize(address_range));
return BAN::move(smo);
const auto& creds = Process::current().credentials();
if (Inode::can_access(info.shm_perm.uid, info.shm_perm.gid, info.shm_perm.mode, creds, flags))
return true;
if (Inode::can_access(info.shm_perm.cuid, info.shm_perm.cgid, info.shm_perm.mode, creds, flags))
return true;
return false;
}
BAN::ErrorOr<BAN::UniqPtr<SharedMemoryObject>> SharedMemoryObject::create(BAN::RefPtr<SharedMemoryObjectManager::Object> object, PageTable& page_table, AddressRange address_range, int status_flags)
{
auto shared_memory_object = TRY(BAN::UniqPtr<SharedMemoryObject>::create(object, page_table, status_flags));
TRY(shared_memory_object->initialize(address_range));
return shared_memory_object;
}
SharedMemoryObject::SharedMemoryObject(BAN::RefPtr<SharedMemoryObjectManager::Object> object, PageTable& page_table, int status_flags)
: MemoryRegion(
page_table,
object->info.shm_segsz,
MemoryRegion::Type::SHARED,
PageTable::UserSupervisor | ((status_flags & O_WRONLY) ? PageTable::ReadWrite : 0) | PageTable::Present,
status_flags
), m_object(object)
{
LockGuard _(m_object->mutex);
m_object->info.shm_nattch++;
m_object->info.shm_atime = SystemTimer::get().real_time().tv_sec;
}
SharedMemoryObject::~SharedMemoryObject()
{
LockGuard _(m_object->mutex);
if (--m_object->info.shm_nattch == 0 && m_object->marked_for_deletion)
SharedMemoryObjectManager::get().m_objects.remove(m_object->key);
m_object->info.shm_dtime = SystemTimer::get().real_time().tv_sec;
}
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> SharedMemoryObject::clone(PageTable& new_page_table)
{
auto region = TRY(SharedMemoryObject::create(m_object, new_page_table, { .start = vaddr(), .end = vaddr() + size() }));
auto region = TRY(SharedMemoryObject::create(m_object, new_page_table, { .start = vaddr(), .end = vaddr() + size() }, m_status_flags));
return BAN::UniqPtr<MemoryRegion>(BAN::move(region));
}
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> SharedMemoryObject::split(size_t offset)
BAN::ErrorOr<BAN::UniqPtr<MemoryRegion>> SharedMemoryObject::split(size_t)
{
(void)offset;
dwarnln("TODO: SharedMemoryObject::split");
return BAN::Error::from_errno(ENOTSUP);
derrorln("SharedMemoryObjects are not splittable");
return BAN::Error::from_errno(EINVAL);
}
BAN::ErrorOr<bool> SharedMemoryObject::allocate_page_containing_impl(vaddr_t address, bool wants_write)
@@ -99,12 +249,11 @@ namespace Kernel
ASSERT(contains(address));
(void)wants_write;
// Check if address is already mapped
vaddr_t vaddr = address & PAGE_ADDR_MASK;
const vaddr_t vaddr = address & PAGE_ADDR_MASK;
if (m_page_table.physical_address_of(vaddr) != 0)
return false;
SpinLockGuard _(m_object->spin_lock);
LockGuard _(m_object->mutex);
paddr_t paddr = m_object->paddrs[(vaddr - m_vaddr) / PAGE_SIZE];
if (paddr == 0)
@@ -117,6 +266,7 @@ namespace Kernel
});
m_object->paddrs[(vaddr - m_vaddr) / PAGE_SIZE] = paddr;
}
m_page_table.map_page_at(paddr, vaddr, m_flags);
return true;
+26 -28
View File
@@ -2928,48 +2928,46 @@ namespace Kernel
return 0;
}
BAN::ErrorOr<long> Process::sys_smo_create(size_t len, int prot)
BAN::ErrorOr<long> Process::sys_shmget(key_t key, size_t size, int shmflg)
{
if (len == 0)
return BAN::Error::from_errno(EINVAL);
if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC | PROT_NONE))
return BAN::Error::from_errno(EINVAL);
if (auto rem = len % PAGE_SIZE)
len += PAGE_SIZE - rem;
PageTable::flags_t page_flags = 0;
if (prot & PROT_READ)
page_flags |= PageTable::Flags::Present;
if (prot & PROT_WRITE)
page_flags |= PageTable::Flags::ReadWrite | PageTable::Flags::Present;
if (prot & PROT_EXEC)
page_flags |= PageTable::Flags::Execute | PageTable::Flags::Present;
if (page_flags == 0)
page_flags |= PageTable::Flags::Reserved;
else
page_flags |= PageTable::Flags::UserSupervisor;
return TRY(SharedMemoryObjectManager::get().create_object(len, page_flags));
return TRY(SharedMemoryObjectManager::get().shmget(key, size, shmflg));
}
BAN::ErrorOr<long> Process::sys_smo_delete(SharedMemoryObjectManager::Key key)
BAN::ErrorOr<long> Process::sys_shmctl(int shmid, int cmd, struct shmid_ds* user_buf)
{
TRY(SharedMemoryObjectManager::get().delete_object(key));
TRY(SharedMemoryObjectManager::get().shmctl(shmid, cmd, user_buf));
return 0;
}
BAN::ErrorOr<long> Process::sys_smo_map(SharedMemoryObjectManager::Key key)
BAN::ErrorOr<long> Process::sys_shmat(int shmid, const void* shmaddr, int shmflg)
{
auto region = TRY(SharedMemoryObjectManager::get().map_object(key, page_table(), { .start = 0x400000, .end = USERSPACE_END }));
RWLockWRGuard _(m_memory_region_lock);
auto region = TRY(SharedMemoryObjectManager::get().shmat(shmid, shmaddr, shmflg));
LockGuard _(m_process_lock);
const vaddr_t region_vaddr = region->vaddr();
TRY(add_mapped_region(BAN::move(region)));
return region_vaddr;
}
BAN::ErrorOr<long> Process::sys_shmdt(const void* shmaddr)
{
const vaddr_t vaddr = reinterpret_cast<vaddr_t>(shmaddr);
RWLockWRGuard _(m_memory_region_lock);
const size_t idx = find_mapped_region(vaddr);
if (idx >= m_mapped_regions.size() || m_mapped_regions[idx]->vaddr() != vaddr)
return BAN::Error::from_errno(ENOENT);
m_mapped_regions[idx]->wait_not_pinned();
m_mapped_regions.remove(idx);
return 0;
}
BAN::ErrorOr<long> Process::sys_ttyname(int fildes, char* user_buffer, size_t buffer_size)
{
auto inode = TRY(m_open_file_descriptors.inode_of(fildes));
+1 -1
View File
@@ -15,7 +15,7 @@
namespace Kernel
{
static_assert(SYS_SIGPROCMASK == 78, "this is hard coded in arch/*/Signal.S");
static_assert(SYS_SIGPROCMASK == 79, "this is hard coded in arch/*/Signal.S");
static_assert(SIG_SETMASK == 3, "this is hard coded in arch/*/Signal.S");
extern "C" [[noreturn]] void start_kernel_thread();