mirror of
https://github.com/LadybirdBrowser/ladybird.git
synced 2024-11-22 15:40:19 +00:00
Kernel: Simplify ProcFS generated buffer caching
Use a Mutex instead of a SpinLock to protect the per-FileDescription generated data cache. This allows processes to go to sleep while waiting their turn. Also don't try to be clever by reusing existing cache buffers. Just allocate KBuffers as needed (and make sure to surface failures.)
This commit is contained in:
parent
4d2473b7fa
commit
a3063dfd33
Notes:
sideshowbarker
2024-07-18 08:40:49 +09:00
Author: https://github.com/awesomekling Commit: https://github.com/SerenityOS/serenity/commit/a3063dfd337
2 changed files with 28 additions and 37 deletions
|
@ -69,7 +69,7 @@ ProcFSExposedLink::ProcFSExposedLink(StringView name, InodeIndex preallocated_in
|
|||
}
|
||||
|
||||
struct ProcFSInodeData : public FileDescriptionData {
|
||||
RefPtr<KBufferImpl> buffer;
|
||||
OwnPtr<KBuffer> buffer;
|
||||
};
|
||||
|
||||
KResultOr<size_t> ProcFSGlobalInformation::read_bytes(off_t offset, size_t count, UserOrKernelBuffer& buffer, FileDescription* description) const
|
||||
|
@ -81,13 +81,16 @@ KResultOr<size_t> ProcFSGlobalInformation::read_bytes(off_t offset, size_t count
|
|||
|
||||
if (!description)
|
||||
return KResult(EIO);
|
||||
|
||||
MutexLocker locker(m_refresh_lock);
|
||||
|
||||
if (!description->data()) {
|
||||
dbgln("ProcFSGlobalInformation: Do not have cached data!");
|
||||
return KResult(EIO);
|
||||
}
|
||||
|
||||
// Be sure to keep a reference to data_buffer while we use it!
|
||||
RefPtr<KBufferImpl> data_buffer = static_cast<ProcFSInodeData&>(*description->data()).buffer;
|
||||
auto& typed_cached_data = static_cast<ProcFSInodeData&>(*description->data());
|
||||
auto& data_buffer = typed_cached_data.buffer;
|
||||
|
||||
if (!data_buffer || (size_t)offset >= data_buffer->size())
|
||||
return 0;
|
||||
|
@ -101,26 +104,19 @@ KResultOr<size_t> ProcFSGlobalInformation::read_bytes(off_t offset, size_t count
|
|||
|
||||
KResult ProcFSGlobalInformation::refresh_data(FileDescription& description) const
|
||||
{
|
||||
ScopedSpinLock lock(m_refresh_lock);
|
||||
MutexLocker lock(m_refresh_lock);
|
||||
auto& cached_data = description.data();
|
||||
if (!cached_data)
|
||||
if (!cached_data) {
|
||||
cached_data = adopt_own_if_nonnull(new (nothrow) ProcFSInodeData);
|
||||
VERIFY(description.data());
|
||||
auto& buffer = static_cast<ProcFSInodeData&>(*cached_data).buffer;
|
||||
if (buffer) {
|
||||
// If we're reusing the buffer, reset the size to 0 first. This
|
||||
// ensures we don't accidentally leak previously written data.
|
||||
buffer->set_size(0);
|
||||
if (!cached_data)
|
||||
return ENOMEM;
|
||||
}
|
||||
KBufferBuilder builder(buffer, true);
|
||||
KBufferBuilder builder;
|
||||
if (!const_cast<ProcFSGlobalInformation&>(*this).output(builder))
|
||||
return ENOENT;
|
||||
// We don't use builder.build() here, which would steal our buffer
|
||||
// and turn it into an OwnPtr. Instead, just flush to the buffer so
|
||||
// that we can read all the data that was written.
|
||||
if (!builder.flush())
|
||||
return ENOMEM;
|
||||
if (!buffer)
|
||||
auto& typed_cached_data = static_cast<ProcFSInodeData&>(*cached_data);
|
||||
typed_cached_data.buffer = builder.build();
|
||||
if (!typed_cached_data.buffer)
|
||||
return ENOMEM;
|
||||
return KSuccess;
|
||||
}
|
||||
|
@ -139,8 +135,10 @@ KResultOr<size_t> ProcFSProcessInformation::read_bytes(off_t offset, size_t coun
|
|||
return KResult(EIO);
|
||||
}
|
||||
|
||||
// Be sure to keep a reference to data_buffer while we use it!
|
||||
RefPtr<KBufferImpl> data_buffer = static_cast<ProcFSInodeData&>(*description->data()).buffer;
|
||||
MutexLocker locker(m_refresh_lock);
|
||||
|
||||
auto& typed_cached_data = static_cast<ProcFSInodeData&>(*description->data());
|
||||
auto& data_buffer = typed_cached_data.buffer;
|
||||
|
||||
if (!data_buffer || (size_t)offset >= data_buffer->size())
|
||||
return 0;
|
||||
|
@ -171,26 +169,19 @@ KResult ProcFSProcessInformation::refresh_data(FileDescription& description) con
|
|||
ScopeGuard guard = [&] {
|
||||
process->ptrace_lock().unlock();
|
||||
};
|
||||
ScopedSpinLock lock(m_refresh_lock);
|
||||
MutexLocker locker(m_refresh_lock);
|
||||
auto& cached_data = description.data();
|
||||
if (!cached_data)
|
||||
if (!cached_data) {
|
||||
cached_data = adopt_own_if_nonnull(new (nothrow) ProcFSInodeData);
|
||||
VERIFY(description.data());
|
||||
auto& buffer = static_cast<ProcFSInodeData&>(*cached_data).buffer;
|
||||
if (buffer) {
|
||||
// If we're reusing the buffer, reset the size to 0 first. This
|
||||
// ensures we don't accidentally leak previously written data.
|
||||
buffer->set_size(0);
|
||||
if (!cached_data)
|
||||
return ENOMEM;
|
||||
}
|
||||
KBufferBuilder builder(buffer, true);
|
||||
KBufferBuilder builder;
|
||||
if (!const_cast<ProcFSProcessInformation&>(*this).output(builder))
|
||||
return ENOENT;
|
||||
// We don't use builder.build() here, which would steal our buffer
|
||||
// and turn it into an OwnPtr. Instead, just flush to the buffer so
|
||||
// that we can read all the data that was written.
|
||||
if (!builder.flush())
|
||||
return ENOMEM;
|
||||
if (!buffer)
|
||||
auto& typed_cached_data = static_cast<ProcFSInodeData&>(*cached_data);
|
||||
typed_cached_data.buffer = builder.build();
|
||||
if (!typed_cached_data.buffer)
|
||||
return ENOMEM;
|
||||
return KSuccess;
|
||||
}
|
||||
|
|
|
@ -187,7 +187,7 @@ protected:
|
|||
virtual KResult refresh_data(FileDescription&) const override;
|
||||
virtual bool output(KBufferBuilder& builder) = 0;
|
||||
|
||||
mutable SpinLock<u8> m_refresh_lock;
|
||||
mutable Mutex m_refresh_lock;
|
||||
};
|
||||
|
||||
class ProcFSSystemBoolean : public ProcFSGlobalInformation {
|
||||
|
@ -245,7 +245,7 @@ protected:
|
|||
virtual bool output(KBufferBuilder& builder) = 0;
|
||||
|
||||
WeakPtr<ProcFSProcessDirectory> m_parent_directory;
|
||||
mutable SpinLock<u8> m_refresh_lock;
|
||||
mutable Mutex m_refresh_lock;
|
||||
};
|
||||
|
||||
}
|
||||
|
|
Loading…
Reference in a new issue