Set of offset-based APIs for thread-safe file IO (#53669)

Co-authored-by: Stephen Toub <stoub@microsoft.com>

src/libraries/Common/src/Interop/Unix/System.Native/Interop.IOVector.cs

@@ -0,0 +1,16 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System;
+
+internal static partial class Interop
+{
+    internal static partial class Sys
+    {
+        internal unsafe struct IOVector
+        {
+            public byte* Base;
+            public UIntPtr Count;
+        }
+    }
+}

src/libraries/Common/src/Interop/Unix/System.Native/Interop.MessageHeader.cs

@@ -8,12 +8,6 @@ internal static partial class Interop
 {
     internal static partial class Sys
     {
-        internal unsafe struct IOVector
-        {
-            public byte* Base;
-            public UIntPtr Count;
-        }
-
         internal unsafe struct MessageHeader
         {
             public byte* SocketAddress;

src/libraries/Common/src/Interop/Unix/System.Native/Interop.PRead.cs

@@ -0,0 +1,13 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System.Runtime.InteropServices;
+
+internal static partial class Interop
+{
+    internal static partial class Sys
+    {
+        [DllImport(Libraries.SystemNative, EntryPoint = "SystemNative_PRead", SetLastError = true)]
+        internal static extern unsafe int PRead(SafeHandle fd, byte* buffer, int bufferSize, long fileOffset);
+    }
+}

src/libraries/Common/src/Interop/Unix/System.Native/Interop.PReadV.cs

@@ -0,0 +1,13 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System.Runtime.InteropServices;
+
+internal static partial class Interop
+{
+    internal static partial class Sys
+    {
+        [DllImport(Libraries.SystemNative, EntryPoint = "SystemNative_PReadV", SetLastError = true)]
+        internal static extern unsafe long PReadV(SafeHandle fd, IOVector* vectors, int vectorCount, long fileOffset);
+    }
+}

src/libraries/Common/src/Interop/Unix/System.Native/Interop.PWrite.cs

@@ -0,0 +1,13 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System.Runtime.InteropServices;
+
+internal static partial class Interop
+{
+    internal static partial class Sys
+    {
+        [DllImport(Libraries.SystemNative, EntryPoint = "SystemNative_PWrite", SetLastError = true)]
+        internal static extern unsafe int PWrite(SafeHandle fd, byte* buffer, int bufferSize, long fileOffset);
+    }
+}

src/libraries/Common/src/Interop/Unix/System.Native/Interop.PWriteV.cs

@@ -0,0 +1,13 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System.Runtime.InteropServices;
+
+internal static partial class Interop
+{
+    internal static partial class Sys
+    {
+        [DllImport(Libraries.SystemNative, EntryPoint = "SystemNative_PWriteV", SetLastError = true)]
+        internal static extern unsafe long PWriteV(SafeHandle fd, IOVector* vectors, int vectorCount, long fileOffset);
+    }
+}

src/libraries/Common/src/Interop/Windows/Kernel32/Interop.FileScatterGather.cs

@@ -0,0 +1,28 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System;
+using System.Runtime.InteropServices;
+using System.Threading;
+
+internal static partial class Interop
+{
+    internal static partial class Kernel32
+    {
+        [DllImport(Libraries.Kernel32, SetLastError = true)]
+        internal static extern unsafe int ReadFileScatter(
+            SafeHandle hFile,
+            long* aSegmentArray,
+            int nNumberOfBytesToRead,
+            IntPtr lpReserved,
+            NativeOverlapped* lpOverlapped);
+
+        [DllImport(Libraries.Kernel32, SetLastError = true)]
+        internal static extern unsafe int WriteFileGather(
+            SafeHandle hFile,
+            long* aSegmentArray,
+            int nNumberOfBytesToWrite,
+            IntPtr lpReserved,
+            NativeOverlapped* lpOverlapped);
+    }
+}

src/libraries/Common/src/Interop/Windows/NtDll/Interop.NtCreateFile.cs

@@ -10,10 +10,6 @@ internal static partial class Interop
 {
     internal static partial class NtDll
     {
-        internal const uint NT_ERROR_STATUS_DISK_FULL = 0xC000007F;
-        internal const uint NT_ERROR_STATUS_FILE_TOO_LARGE = 0xC0000904;
-        internal const uint NT_STATUS_INVALID_PARAMETER = 0xC000000D;
-
         // https://msdn.microsoft.com/en-us/library/bb432380.aspx
         // https://msdn.microsoft.com/en-us/library/windows/hardware/ff566424.aspx
         [DllImport(Libraries.NtDll, CharSet = CharSet.Unicode, ExactSpelling = true)]
@@ -76,7 +72,7 @@ internal static unsafe (uint status, IntPtr handle) CreateFile(
             }
         }
 
-        internal static unsafe (uint status, IntPtr handle) CreateFile(ReadOnlySpan<char> path, FileMode mode, FileAccess access, FileShare share, FileOptions options, long preallocationSize)
+        internal static unsafe (uint status, IntPtr handle) NtCreateFile(ReadOnlySpan<char> path, FileMode mode, FileAccess access, FileShare share, FileOptions options, long preallocationSize)
         {
             // For mitigating local elevation of privilege attack through named pipes
             // make sure we always call NtCreateFile with SECURITY_ANONYMOUS so that the
@@ -120,7 +116,7 @@ private static CreateDisposition GetCreateDisposition(FileMode mode)
 
         private static DesiredAccess GetDesiredAccess(FileAccess access, FileMode fileMode, FileOptions options)
         {
-            DesiredAccess result = 0;
+            DesiredAccess result = DesiredAccess.FILE_READ_ATTRIBUTES | DesiredAccess.SYNCHRONIZE; // default values used by CreateFileW
 
             if ((access & FileAccess.Read) != 0)
             {
@@ -134,13 +130,9 @@ private static DesiredAccess GetDesiredAccess(FileAccess access, FileMode fileMo
             {
                 result |= DesiredAccess.FILE_APPEND_DATA;
             }
-            if ((options & FileOptions.Asynchronous) == 0)
-            {
-                result |= DesiredAccess.SYNCHRONIZE; // required by FILE_SYNCHRONOUS_IO_NONALERT
-            }
-            if ((options & FileOptions.DeleteOnClose) != 0 || fileMode == FileMode.Create)
+            if ((options & FileOptions.DeleteOnClose) != 0)
             {
-                result |= DesiredAccess.DELETE; // required by FILE_DELETE_ON_CLOSE and FILE_SUPERSEDE (which deletes a file if it exists)
+                result |= DesiredAccess.DELETE; // required by FILE_DELETE_ON_CLOSE
             }
 
             return result;
@@ -190,7 +182,8 @@ private static CreateOptions GetCreateOptions(FileOptions options)
         }
 
         private static ObjectAttributes GetObjectAttributes(FileShare share)
-            => (share & FileShare.Inheritable) != 0 ? ObjectAttributes.OBJ_INHERIT : 0;
+            => ObjectAttributes.OBJ_CASE_INSENSITIVE | // default value used by CreateFileW
+                ((share & FileShare.Inheritable) != 0 ? ObjectAttributes.OBJ_INHERIT : 0);
 
         /// <summary>
         /// File creation disposition when calling directly to NT APIs.

src/libraries/Common/src/Interop/Windows/NtDll/Interop.NtQueryInformationFile.cs

@@ -2,7 +2,6 @@
 // The .NET Foundation licenses this file to you under the MIT license.
 
 using Microsoft.Win32.SafeHandles;
-using System;
 using System.Runtime.InteropServices;
 
 internal static partial class Interop

src/libraries/Common/src/Interop/Windows/NtDll/Interop.NtStatus.cs

@@ -17,5 +17,7 @@ internal static class StatusOptions
         internal const uint STATUS_ACCOUNT_RESTRICTION    = 0xC000006E;
         internal const uint STATUS_NONE_MAPPED            = 0xC0000073;
         internal const uint STATUS_INSUFFICIENT_RESOURCES = 0xC000009A;
+        internal const uint STATUS_DISK_FULL = 0xC000007F;
+        internal const uint STATUS_FILE_TOO_LARGE = 0xC0000904;
     }
 }

src/libraries/Native/Unix/Common/pal_config.h.in

@@ -36,6 +36,8 @@
 #cmakedefine01 HAVE_POSIX_ADVISE
 #cmakedefine01 HAVE_POSIX_FALLOCATE
 #cmakedefine01 HAVE_POSIX_FALLOCATE64
+#cmakedefine01 HAVE_PREADV
+#cmakedefine01 HAVE_PWRITEV
 #cmakedefine01 PRIORITY_REQUIRES_INT_WHO
 #cmakedefine01 KEVENT_REQUIRES_INT_PARAMS
 #cmakedefine01 HAVE_IOCTL

src/libraries/Native/Unix/System.Native/entrypoints.c

@@ -242,6 +242,10 @@ static const Entry s_sysNative[] =
     DllImportEntry(SystemNative_iOSSupportVersion)
     DllImportEntry(SystemNative_GetErrNo)
     DllImportEntry(SystemNative_SetErrNo)
+    DllImportEntry(SystemNative_PRead)
+    DllImportEntry(SystemNative_PWrite)
+    DllImportEntry(SystemNative_PReadV)
+    DllImportEntry(SystemNative_PWriteV)
 };
 
 EXTERN_C const void* SystemResolveDllImport(const char* name);

src/libraries/Native/Unix/System.Native/pal_io.c

@@ -22,6 +22,7 @@
 #include <sys/file.h>
 #include <sys/ioctl.h>
 #include <sys/socket.h>
+#include <sys/uio.h>
 #include <syslog.h>
 #include <termios.h>
 #include <unistd.h>
@@ -1454,3 +1455,107 @@ int32_t SystemNative_ReadProcessStatusInfo(pid_t pid, ProcessStatus* processStat
     return -1;
 #endif // __sun
 }
+
+int32_t SystemNative_PRead(intptr_t fd, void* buffer, int32_t bufferSize, int64_t fileOffset)
+{
+    assert(buffer != NULL);
+    assert(bufferSize >= 0);
+
+    ssize_t count;
+    while ((count = pread(ToFileDescriptor(fd), buffer, (uint32_t)bufferSize, (off_t)fileOffset)) < 0 && errno == EINTR);
+
+    assert(count >= -1 && count <= bufferSize);
+    return (int32_t)count;
+}
+
+int32_t SystemNative_PWrite(intptr_t fd, void* buffer, int32_t bufferSize, int64_t fileOffset)
+{
+    assert(buffer != NULL);
+    assert(bufferSize >= 0);
+
+    ssize_t count;
+    while ((count = pwrite(ToFileDescriptor(fd), buffer, (uint32_t)bufferSize, (off_t)fileOffset)) < 0 && errno == EINTR);
+
+    assert(count >= -1 && count <= bufferSize);
+    return (int32_t)count;
+}
+
+int64_t SystemNative_PReadV(intptr_t fd, IOVector* vectors, int32_t vectorCount, int64_t fileOffset)
+{
+    assert(vectors != NULL);
+    assert(vectorCount >= 0);
+
+    int64_t count = 0;
+    int fileDescriptor = ToFileDescriptor(fd);
+#if HAVE_PREADV && !defined(TARGET_WASM) // preadv is buggy on WASM
+    while ((count = preadv(fileDescriptor, (struct iovec*)vectors, (int)vectorCount, (off_t)fileOffset)) < 0 && errno == EINTR);
+#else
+    int64_t current;
+    for (int i = 0; i < vectorCount; i++)
+    {
+        IOVector vector = vectors[i];
+        while ((current = pread(fileDescriptor, vector.Base, vector.Count, (off_t)(fileOffset + count))) < 0 && errno == EINTR);
+
+        if (current < 0)
+        {
+            // if previous calls were succesfull, we return what we got so far
+            // otherwise, we return the error code
+            return count > 0 ? count : current;
+        }
+
+        count += current;
+
+        // Incomplete pread operation may happen for two reasons:
+        // a) We have reached EOF.
+        // b) The operation was interrupted by a signal handler.
+        // To mimic preadv, we stop on the first incomplete operation.
+        if (current != (int64_t)vector.Count)
+        {
+            return count;
+        }
+    }
+#endif
+
+    assert(count >= -1);
+    return count;
+}
+
+int64_t SystemNative_PWriteV(intptr_t fd, IOVector* vectors, int32_t vectorCount, int64_t fileOffset)
+{
+    assert(vectors != NULL);
+    assert(vectorCount >= 0);
+
+    int64_t count = 0;
+    int fileDescriptor = ToFileDescriptor(fd);
+#if HAVE_PWRITEV && !defined(TARGET_WASM) // pwritev is buggy on WASM
+    while ((count = pwritev(fileDescriptor, (struct iovec*)vectors, (int)vectorCount, (off_t)fileOffset)) < 0 && errno == EINTR);
+#else
+    int64_t current;
+    for (int i = 0; i < vectorCount; i++)
+    {
+        IOVector vector = vectors[i];
+        while ((current = pwrite(fileDescriptor, vector.Base, vector.Count, (off_t)(fileOffset + count))) < 0 && errno == EINTR);
+
+        if (current < 0)
+        {
+            // if previous calls were succesfull, we return what we got so far
+            // otherwise, we return the error code
+            return count > 0 ? count : current;
+        }
+
+        count += current;
+
+        // Incomplete pwrite operation may happen for few reasons:
+        // a) There was not enough space available or the file is too large for given file system.
+        // b) The operation was interrupted by a signal handler.
+        // To mimic pwritev, we stop on the first incomplete operation.
+        if (current != (int64_t)vector.Count)
+        {
+            return count;
+        }
+    }
+#endif
+
+    assert(count >= -1);
+    return count;
+}

src/libraries/Native/Unix/System.Native/pal_io.h

@@ -41,6 +41,14 @@ typedef struct
     // add more fields when needed.
 } ProcessStatus;
 
+// NOTE: the layout of this type is intended to exactly  match the layout of a `struct iovec`. There are
+//       assertions in pal_networking.c that validate this.
+typedef struct
+{
+    uint8_t* Base;
+    uintptr_t Count;
+} IOVector;
+
 /* Provide consistent access to nanosecond fields, if they exist. */
 /* Seconds are always available through st_atime, st_mtime, st_ctime. */
 
@@ -730,3 +738,31 @@ PALEXPORT int32_t SystemNative_LChflagsCanSetHiddenFlag(void);
  * Returns 1 if the process status was read; otherwise, 0.
  */
 PALEXPORT int32_t SystemNative_ReadProcessStatusInfo(pid_t pid, ProcessStatus* processStatus);
+
+/**
+ * Reads the number of bytes specified into the provided buffer from the specified, opened file descriptor at specified offset.
+ *
+ * Returns the number of bytes read on success; otherwise, -1 is returned an errno is set.
+ */
+PALEXPORT int32_t SystemNative_PRead(intptr_t fd, void* buffer, int32_t bufferSize, int64_t fileOffset);
+
+/**
+ * Writes the number of bytes specified in the buffer into the specified, opened file descriptor at specified offset.
+ *
+ * Returns the number of bytes written on success; otherwise, -1 is returned an errno is set.
+ */
+PALEXPORT int32_t SystemNative_PWrite(intptr_t fd, void* buffer, int32_t bufferSize, int64_t fileOffset);
+
+/**
+ * Reads the number of bytes specified into the provided buffers from the specified, opened file descriptor at specified offset.
+ *
+ * Returns the number of bytes read on success; otherwise, -1 is returned an errno is set.
+ */
+PALEXPORT int64_t SystemNative_PReadV(intptr_t fd, IOVector* vectors, int32_t vectorCount, int64_t fileOffset);
+
+/**
+ * Writes the number of bytes specified in the buffers into the specified, opened file descriptor at specified offset.
+ *
+ * Returns the number of bytes written on success; otherwise, -1 is returned an errno is set.
+ */
+PALEXPORT int64_t SystemNative_PWriteV(intptr_t fd, IOVector* vectors, int32_t vectorCount, int64_t fileOffset);

src/libraries/Native/Unix/System.Native/pal_networking.c

@@ -3,7 +3,6 @@
 
 #include "pal_config.h"
 #include "pal_networking.h"
-#include "pal_io.h"
 #include "pal_safecrt.h"
 #include "pal_utilities.h"
 #include <pal_networking_common.h>
@@ -3104,11 +3103,11 @@ int32_t SystemNative_Disconnect(intptr_t socket)
     addr.sa_family = AF_UNSPEC;
 
     err = connect(fd, &addr, sizeof(addr));
-    if (err != 0) 
+    if (err != 0)
     {
         // On some older kernels connect(AF_UNSPEC) may fail. Fall back to shutdown in these cases:
         err = shutdown(fd, SHUT_RDWR);
-    } 
+    }
 #elif HAVE_DISCONNECTX
     // disconnectx causes a FIN close on OSX. It's the best we can do.
     err = disconnectx(fd, SAE_ASSOCID_ANY, SAE_CONNID_ANY);