[PATCH v16 0/6] MR5769: Add support for large page backed file mapping objects on Linux.
This MR adds support for creating file mapping objects backed by large pages on Linux, by making the following changes:
## wineserver
* On Linux, `create_temp_file` will first attempt to use memfds as the backing fd. If it fails, it'll return to the current codepath, creating a temporary file in either the server or config directory. * The created memfd will be sealed against writes, if the caller requesting the appropriate page protection flags. * This removes the requirement that FDs be only created on filesystems/directories that aren't `noexec`. * In the server method `create_mapping` , if large pages have been requested by the caller, hold that the calling thread's token holds `SeLockMemoryPrivilege` . * Additionally, add `SeLockMemoryPrivilege` to the list of privileges enabled for the Administrator.
## `ntdll`
* Add `virtual_get_min_large_page_size` and its exported wrapper `wine_unix_get_min_large_page_size`. * On Linux, the minimum page size is determined by going through `/sys/kernel/mm/hugepages`. If hugepage support was not detected, `STATUS_NOT_SUPPORTED` is returned instead. On other platforms, the older hard-coded value of 2*1024*1024 is returned instead. * `NtCreateSection` will validate certain parameters if large pages are requested. Specifically, it will return STATUS_INVALID_PARAMETER if the requested mapping is not anonymous/unnamed, or the size is not a multiple of the minimum supported page size.
## `kernelbase`
* `GetLargePageMinimum` will use `wine_unix_get_min_large_page_size`.
## `kernel32/tests`
* Add new test test_large_page_file_mapping, which validates privilege enforcements and parameter validation while creating large pages backed file mapping obejcts. The tests are skipped if `GetLargePageMinimum` returns 0.
-- v16: kernel32: Add tests for large page mapping support. ntdll: Validate parameters to NtCreateSection if large pages are requested. server: Use memfd to back anonymous mappings on Linux. kernelbase: Implement GetLargePageMinimum by returning the value of LargePageMinimum in _KUSER_SHARED_DATA. server: Set LargePageMinimum in _KUSER_SHARED_DATA on Linux. server: Require SeLockMemoryPrivilege to create large page mappings.
From: Vibhav Pant vibhavp@gmail.com
Also, add SeLockMemoryPrivilege to the list of admin privileges. --- server/mapping.c | 6 ++++++ server/security.h | 1 + server/token.c | 2 ++ 3 files changed, 9 insertions(+)
diff --git a/server/mapping.c b/server/mapping.c index ff99b45ce51..a6f7d3e951e 100644 --- a/server/mapping.c +++ b/server/mapping.c @@ -1284,6 +1284,12 @@ DECL_HANDLER(create_mapping)
if (!objattr) return;
+ if ((req->flags & SEC_LARGE_PAGES) && !thread_single_check_privilege( current, SeLockMemoryPrivilege )) + { + set_error( STATUS_PRIVILEGE_NOT_HELD ); + return; + } + if ((mapping = create_mapping( root, &name, objattr->attributes, req->size, req->flags, req->file_handle, req->file_access, sd ))) { diff --git a/server/security.h b/server/security.h index f4dff679179..7a135e90928 100644 --- a/server/security.h +++ b/server/security.h @@ -23,6 +23,7 @@
#include <sys/types.h>
+extern const struct luid SeLockMemoryPrivilege; extern const struct luid SeIncreaseQuotaPrivilege; extern const struct luid SeSecurityPrivilege; extern const struct luid SeTakeOwnershipPrivilege; diff --git a/server/token.c b/server/token.c index da7f0bb7ff2..b1fd23af492 100644 --- a/server/token.c +++ b/server/token.c @@ -42,6 +42,7 @@
#define MAX_SUBAUTH_COUNT 1
+const struct luid SeLockMemoryPrivilege = { 4, 0}; const struct luid SeIncreaseQuotaPrivilege = { 5, 0 }; const struct luid SeTcbPrivilege = { 7, 0 }; const struct luid SeSecurityPrivilege = { 8, 0 }; @@ -782,6 +783,7 @@ struct token *token_create_admin( unsigned primary, int impersonation_level, int { SeManageVolumePrivilege, 0 }, { SeImpersonatePrivilege, SE_PRIVILEGE_ENABLED }, { SeCreateGlobalPrivilege, SE_PRIVILEGE_ENABLED }, + { SeLockMemoryPrivilege, SE_PRIVILEGE_ENABLED }, }; /* note: we don't include non-builtin groups here for the user - * telling us these is the job of a client-side program */
From: Vibhav Pant vibhavp@gmail.com
On Linux, get the minimum supported hugepage size from directory entries /sys/kernel/mm/hugepages, and set the LargePageMinimum field in user_shared_data to it, if supported. --- server/mapping.c | 75 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 75 insertions(+)
diff --git a/server/mapping.c b/server/mapping.c index a6f7d3e951e..04cd60a1d01 100644 --- a/server/mapping.c +++ b/server/mapping.c @@ -29,6 +29,14 @@ #include <sys/stat.h> #include <sys/mman.h> #include <unistd.h> +#ifdef HAVE_MEMFD_CREATE +#include <linux/memfd.h> +#endif +#ifdef __linux__ /* For detecting hugepage support. */ +#include <dirent.h> +#include <errno.h> +#include <limits.h> +#endif
#include "ntstatus.h" #define WIN32_NO_STATUS @@ -1256,6 +1264,66 @@ int get_page_size(void) return page_mask + 1; }
+#ifdef __linux__ +static size_t linux_get_min_hugepage_size( void ) +{ + DIR *sysfs_hugepages; + struct dirent *supported_size; + size_t min_size = 0; + size_t total_supported_sizes = 0; + + errno = 0; + sysfs_hugepages = opendir("/sys/kernel/mm/hugepages"); + if (sysfs_hugepages == NULL) + { + return -1; + } + + while(1) + { + long hugepage_size; + + supported_size = readdir( sysfs_hugepages ); + if ( supported_size == NULL ) + { + break; + } + if (strncmp( supported_size->d_name, "hugepages-", 10) != 0) + { + continue; + } + errno = 0; + hugepage_size = strtol( &supported_size->d_name[10], NULL, 10 ); + if (hugepage_size == 0 || hugepage_size == LONG_MAX || hugepage_size == LONG_MIN) + { + if (errno != 0) + { + fprintf( stderr, + "could not parse page size from directory entry '%s': %s\n", + supported_size->d_name, strerror( errno ) ); + } + continue; + } + hugepage_size *= 1024; + min_size = ( total_supported_sizes == 0 ) + ? hugepage_size + : ( hugepage_size < min_size ? hugepage_size : min_size ); + total_supported_sizes++; + } + + closedir(sysfs_hugepages); + return min_size; +} +#endif /* __linux__ */ + +static size_t get_min_large_page_size( void ) +{ +#ifdef __linux__ + return linux_get_min_hugepage_size(); +#endif + return 0; +} + struct object *create_user_data_mapping( struct object *root, const struct unicode_str *name, unsigned int attr, const struct security_descriptor *sd ) { @@ -1267,8 +1335,15 @@ struct object *create_user_data_mapping( struct object *root, const struct unico ptr = mmap( NULL, mapping->size, PROT_WRITE, MAP_SHARED, get_unix_fd( mapping->fd ), 0 ); if (ptr != MAP_FAILED) { + ULONG min_large_page_size; + user_shared_data = ptr; user_shared_data->SystemCall = 1; + min_large_page_size = get_min_large_page_size(); + if (min_large_page_size != 0) + { + user_shared_data->LargePageMinimum = min_large_page_size; + } } return &mapping->obj; }
From: Vibhav Pant vibhavp@gmail.com
--- dlls/kernelbase/memory.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/dlls/kernelbase/memory.c b/dlls/kernelbase/memory.c index 4f4bba9a13b..316aec7d40a 100644 --- a/dlls/kernelbase/memory.c +++ b/dlls/kernelbase/memory.c @@ -41,6 +41,7 @@ WINE_DEFAULT_DEBUG_CHANNEL(heap); WINE_DECLARE_DEBUG_CHANNEL(virtual); WINE_DECLARE_DEBUG_CHANNEL(globalmem);
+static const struct _KUSER_SHARED_DATA *user_shared_data = (struct _KUSER_SHARED_DATA *)0x7ffe0000;
/*********************************************************************** * Virtual memory functions @@ -77,7 +78,7 @@ BOOL WINAPI DECLSPEC_HOTPATCH FlushViewOfFile( const void *base, SIZE_T size ) */ SIZE_T WINAPI GetLargePageMinimum(void) { - return 2 * 1024 * 1024; + return user_shared_data->LargePageMinimum; }
From: Vibhav Pant vibhavp@gmail.com
On Linux, try creating a memfd first to implement create_temp_file. If a large page mapping is requested (SEC_LARGE_PAGES), create a memfd with the MFD_HUGETLB flags. If SEC_COMMIT is passed, allocate sufficient space to back the mapping with posix_fallocate.
Additionally, ensure when SEC_LARGE_PAGES is requested, ensure that SEC_COMMIT is passed as well. --- configure | 12 +++++ configure.ac | 2 + include/config.h.in | 6 +++ server/mapping.c | 118 ++++++++++++++++++++++++++++++++++++++++++-- 4 files changed, 134 insertions(+), 4 deletions(-)
diff --git a/configure b/configure index 101981ad8df..d9d4a51c054 100755 --- a/configure +++ b/configure @@ -7995,6 +7995,12 @@ if test "x$ac_cv_header_linux_major_h" = xyes then : printf "%s\n" "#define HAVE_LINUX_MAJOR_H 1" >>confdefs.h
+fi +ac_fn_c_check_header_compile "$LINENO" "linux/memfd.h" "ac_cv_header_linux_memfd_h" "$ac_includes_default" +if test "x$ac_cv_header_linux_memfd_h" = xyes +then : + printf "%s\n" "#define HAVE_LINUX_MEMFD_H 1" >>confdefs.h + fi ac_fn_c_check_header_compile "$LINENO" "linux/param.h" "ac_cv_header_linux_param_h" "$ac_includes_default" if test "x$ac_cv_header_linux_param_h" = xyes @@ -20787,6 +20793,12 @@ if test "x$ac_cv_func_mach_continuous_time" = xyes then : printf "%s\n" "#define HAVE_MACH_CONTINUOUS_TIME 1" >>confdefs.h
+fi +ac_fn_c_check_func "$LINENO" "memfd_create" "ac_cv_func_memfd_create" +if test "x$ac_cv_func_memfd_create" = xyes +then : + printf "%s\n" "#define HAVE_MEMFD_CREATE 1" >>confdefs.h + fi ac_fn_c_check_func "$LINENO" "pipe2" "ac_cv_func_pipe2" if test "x$ac_cv_func_pipe2" = xyes diff --git a/configure.ac b/configure.ac index 231ec7f17d6..e63c97900b2 100644 --- a/configure.ac +++ b/configure.ac @@ -384,6 +384,7 @@ AC_CHECK_HEADERS(\ linux/input.h \ linux/ioctl.h \ linux/major.h \ + linux/memfd.h \ linux/param.h \ linux/serial.h \ linux/types.h \ @@ -2079,6 +2080,7 @@ AC_CHECK_FUNCS(\ getrandom \ kqueue \ mach_continuous_time \ + memfd_create \ pipe2 \ port_create \ posix_fadvise \ diff --git a/include/config.h.in b/include/config.h.in index ef783c75a42..923d943de01 100644 --- a/include/config.h.in +++ b/include/config.h.in @@ -180,6 +180,9 @@ /* Define to 1 if you have the <linux/major.h> header file. */ #undef HAVE_LINUX_MAJOR_H
+/* Define to 1 if you have the <linux/memfd.h> header file. */ +#undef HAVE_LINUX_MEMFD_H + /* Define to 1 if you have the <linux/param.h> header file. */ #undef HAVE_LINUX_PARAM_H
@@ -219,6 +222,9 @@ /* Define to 1 if you have the <mach-o/loader.h> header file. */ #undef HAVE_MACH_O_LOADER_H
+/* Define to 1 if you have the 'memfd_create' function. */ +#undef HAVE_MEMFD_CREATE + /* Define to 1 if you have the <mntent.h> header file. */ #undef HAVE_MNTENT_H
diff --git a/server/mapping.c b/server/mapping.c index 04cd60a1d01..1cf018db629 100644 --- a/server/mapping.c +++ b/server/mapping.c @@ -325,13 +325,117 @@ static int check_current_dir_for_exec(void) return (ret != MAP_FAILED); }
+#ifdef HAVE_MEMFD_CREATE +static void make_memfd_name( const char *prefix, ULONG sec_flags, char *dest ) +{ + /* memfd names are restricted to 250 bytes including \0, but are only meant for debugging, with + * duplicate names not causing any side-effects. */ + + if (prefix != NULL) + { + snprintf( dest, 250, "map-%s-%x", prefix, (int)sec_flags ); + } + else + { + snprintf( dest, 250, "map-%x", (int)sec_flags ); + } +} + +static int create_memfd( ULONG file_access, ULONG sec_flags, file_pos_t new_size, int *err ) +{ + int fd; + off_t size = new_size; + char memfd_name[256]; + unsigned int memfd_flags = MFD_ALLOW_SEALING; + unsigned int seal_flags = F_SEAL_SEAL | F_SEAL_WRITE; + +#ifdef HAVE_POSIX_FALLOCATE + if (sizeof(file_pos_t) > sizeof(off_t) && size != new_size) + { + *err = EINVAL; + return -1; + } +#endif +#ifdef MFD_EXEC + memfd_flags |= MFD_EXEC; +#endif + + make_memfd_name( NULL, sec_flags, memfd_name ); + + if (sec_flags & SEC_LARGE_PAGES) + { + memfd_flags |= MFD_HUGETLB; + } + fd = memfd_create( memfd_name, memfd_flags ); + if (fd == -1) + { + if (errno == EINVAL && (sec_flags & SEC_LARGE_PAGES)) + { + /* MFD_HUGETLB & MFD_ALLOW_SEALING is only available in Linux >= 4.16. Lets try creating + * one without HUGETLB */ + fd = memfd_create( memfd_name, memfd_flags & ~MFD_HUGETLB ); + if (fd == -1) + { + *err = errno; + return -1; + } + } + else + { + *err = errno; + return -1; + } + } +#ifdef HAVE_POSIX_FALLOCATE + if (sec_flags & SEC_COMMIT) + { + *err = posix_fallocate( fd, 0, size ); + if (*err != 0) + { + close( fd ); + return -1; + } + } +#endif + if (ftruncate( fd, size ) == -1) + { + *err = errno; + close( fd ); + return -1; + } + if (file_access & FILE_WRITE_DATA) + { + seal_flags &= ~F_SEAL_WRITE; + } + + if (fcntl( fd, F_ADD_SEALS, F_SEAL_SEAL, seal_flags ) == -1) + { + *err = errno; + close( fd ); + return -1; + } + return fd; +} +#endif /* defined(HAVE_MEMFD_CREATE) */ + /* create a temp file for anonymous mappings */ -static int create_temp_file( file_pos_t size ) +static int create_temp_file( ULONG file_access, ULONG sec_flags, file_pos_t size ) { static int temp_dir_fd = -1; char tmpfn[16]; int fd;
+#ifdef HAVE_MEMFD_CREATE + int err = 0; + + fd = create_memfd( file_access, sec_flags, size, &err); + if (fd != -1) + { + return fd; + } + +#endif + if (temp_dir_fd == -1) { temp_dir_fd = server_dir_fd; @@ -616,7 +720,7 @@ static int build_shared_mapping( struct mapping *mapping, int fd,
/* create a temp file for the mapping */
- if ((shared_fd = create_temp_file( total_size )) == -1) return 0; + if ((shared_fd = create_temp_file( FILE_WRITE_DATA, 0, total_size )) == -1) return 0; if (!(file = create_file_for_fd( shared_fd, FILE_GENERIC_READ|FILE_GENERIC_WRITE, 0 ))) return 0;
if (!(buffer = malloc( max_size ))) goto error; @@ -948,7 +1052,12 @@ static struct ranges *create_ranges(void)
static unsigned int get_mapping_flags( obj_handle_t handle, unsigned int flags ) { - switch (flags & (SEC_IMAGE | SEC_RESERVE | SEC_COMMIT | SEC_FILE)) + if (flags & SEC_LARGE_PAGES) + { + if (!(flags & SEC_COMMIT)) + goto error; + } + switch (flags & (SEC_IMAGE | SEC_RESERVE | SEC_COMMIT | SEC_FILE )) { case SEC_IMAGE: if (flags & (SEC_WRITECOMBINE | SEC_LARGE_PAGES)) break; @@ -963,6 +1072,7 @@ static unsigned int get_mapping_flags( obj_handle_t handle, unsigned int flags ) if (handle) return SEC_FILE | (flags & (SEC_NOCACHE | SEC_WRITECOMBINE)); return flags; } +error: set_error( STATUS_INVALID_PARAMETER ); return 0; } @@ -1052,7 +1162,7 @@ static struct mapping *create_mapping( struct object *root, const struct unicode } if ((flags & SEC_RESERVE) && !(mapping->committed = create_ranges())) goto error; mapping->size = (mapping->size + page_mask) & ~((mem_size_t)page_mask); - if ((unix_fd = create_temp_file( mapping->size )) == -1) goto error; + if ((unix_fd = create_temp_file( file_access, flags, mapping->size )) == -1) goto error; if (!(mapping->fd = create_anonymous_fd( &mapping_fd_ops, unix_fd, &mapping->obj, FILE_SYNCHRONOUS_IO_NONALERT ))) goto error; allow_fd_caching( mapping->fd );
From: Vibhav Pant vibhavp@gmail.com
If large pages are requested (SEC_LARGE_PAGES), return STATUS_INVALID_PARAMETER if: * If the backing file object (file) is not NULL. * The maximum size of the section is NULL. * Large pages are not supported by the host platform. * Large pages are supported, but the requested size is not a multiple of the minimum large page size. --- dlls/ntdll/unix/sync.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
diff --git a/dlls/ntdll/unix/sync.c b/dlls/ntdll/unix/sync.c index 4b2d7c1ccbc..81931931d9c 100644 --- a/dlls/ntdll/unix/sync.c +++ b/dlls/ntdll/unix/sync.c @@ -2058,6 +2058,19 @@ NTSTATUS WINAPI NtCreateSection( HANDLE *handle, ACCESS_MASK access, const OBJEC
*handle = 0;
+ if (sec_flags & SEC_LARGE_PAGES) + { + SIZE_T min_size = user_shared_data->LargePageMinimum; + if (file != NULL || size == NULL) + { + return STATUS_INVALID_PARAMETER; + } + + if (min_size == 0 || size->QuadPart % min_size != 0) + { + return STATUS_INVALID_PARAMETER; + } + } switch (protect & 0xff) { case PAGE_READONLY:
From: Vibhav Pant vibhavp@gmail.com
--- dlls/kernel32/tests/process.c | 73 +++++++++++++++++++++++++++++++++++ 1 file changed, 73 insertions(+)
diff --git a/dlls/kernel32/tests/process.c b/dlls/kernel32/tests/process.c index 8e168622c17..2521c52a7b1 100644 --- a/dlls/kernel32/tests/process.c +++ b/dlls/kernel32/tests/process.c @@ -82,6 +82,7 @@ static BOOL (WINAPI *pGetNumaProcessorNode)(UCHAR, PUCHAR); static NTSTATUS (WINAPI *pNtQueryInformationProcess)(HANDLE, PROCESSINFOCLASS, PVOID, ULONG, PULONG); static NTSTATUS (WINAPI *pNtQueryInformationThread)(HANDLE, THREADINFOCLASS, PVOID, ULONG, PULONG); static NTSTATUS (WINAPI *pNtQuerySystemInformationEx)(SYSTEM_INFORMATION_CLASS, void*, ULONG, void*, ULONG, ULONG*); +static NTSTATUS (WINAPI *pRtlAdjustPrivilege)(ULONG,BOOLEAN,BOOLEAN,PBOOLEAN); static DWORD (WINAPI *pWTSGetActiveConsoleSessionId)(void); static HANDLE (WINAPI *pCreateToolhelp32Snapshot)(DWORD, DWORD); static BOOL (WINAPI *pProcess32First)(HANDLE, PROCESSENTRY32*); @@ -252,6 +253,7 @@ static BOOL init(void) pNtQueryInformationProcess = (void *)GetProcAddress(hntdll, "NtQueryInformationProcess"); pNtQueryInformationThread = (void *)GetProcAddress(hntdll, "NtQueryInformationThread"); pNtQuerySystemInformationEx = (void *)GetProcAddress(hntdll, "NtQuerySystemInformationEx"); + pRtlAdjustPrivilege = (void *)GetProcAddress(hntdll, "RtlAdjustPrivilege");
pGetNativeSystemInfo = (void *) GetProcAddress(hkernel32, "GetNativeSystemInfo"); pGetSystemRegistryQuota = (void *) GetProcAddress(hkernel32, "GetSystemRegistryQuota"); @@ -4347,6 +4349,76 @@ static void test_largepages(void) ok((size == 0) || (size == 2*1024*1024) || (size == 4*1024*1024), "GetLargePageMinimum reports %Id size\n", size); }
+static void test_large_page_file_mapping( void ) +{ + SIZE_T size; + BOOLEAN enabled; + NTSTATUS status; + DWORD err; + HANDLE file; + HANDLE token; + HANDLE process_token; + + if (!pGetLargePageMinimum) + { + win_skip( "No GetLargePageMinimum support.\n" ); + return; + } + if (!pRtlAdjustPrivilege) + { + win_skip( "No RtlAdjustPrivilege support.\n" ); + return; + } + + size = pGetLargePageMinimum(); + if (size == 0) + { + trace( "No large pages support, skipping test.\n" ); + return; + } + ok( OpenProcessToken( GetCurrentProcess(), TOKEN_DUPLICATE | TOKEN_QUERY, &process_token ), + "OpenProcessToken failed (%ld)\n", GetLastError() ); + ok( DuplicateToken( process_token, SecurityImpersonation, &token ), + "DuplicateToken failed (%ld)\n", GetLastError() ); + ok( ImpersonateLoggedOnUser( token ), "ImpersonateLoggedOnUser failed (%ld)\n", + GetLastError() ); + status = pRtlAdjustPrivilege( SE_LOCK_MEMORY_PRIVILEGE, TRUE, TRUE, &enabled ); + if (status != STATUS_SUCCESS) + { + trace( "Couldn't get SE_LOCK_MEMORY_PRIVILEGE (%ld), skipping large page file " + "mapping test.\n", + status ); + } + else + { + file = + CreateFileMappingW( INVALID_HANDLE_VALUE, NULL, + PAGE_READWRITE | SEC_LARGE_PAGES | SEC_COMMIT, 0, size - 1, NULL ); + err = GetLastError(); + ok( file == NULL && err == ERROR_INVALID_PARAMETER, + "CreateFileMappingW should fail with ERROR_INVALID_PARAMETER (got %ld instead)\n", + err ); + if (file != NULL) CloseHandle( file ); + + file = + CreateFileMappingW( INVALID_HANDLE_VALUE, NULL, + PAGE_READWRITE | SEC_LARGE_PAGES | SEC_COMMIT, 0, size * 2, NULL ); + ok( file != NULL, "CreateFileMappingW failed (%ld)\n", GetLastError() ); + if (file != NULL) CloseHandle( file ); + + file = CreateFileMappingW( INVALID_HANDLE_VALUE, NULL, + PAGE_READWRITE | SEC_LARGE_PAGES | SEC_RESERVE, 0, size, NULL ); + err = GetLastError(); + ok( file == NULL && err == ERROR_INVALID_PARAMETER, + "CreateFileMappingW should have failed with ERROR_INVALID_PARAMETER (got %ld " + "instead)\n", + err ); + if (file != NULL) CloseHandle( file ); + } + + ok( RevertToSelf(), "RevertToSelf failed (%ld)\n", GetLastError() ); +} + struct proc_thread_attr { DWORD_PTR attr; @@ -5668,6 +5740,7 @@ START_TEST(process) test_GetLogicalProcessorInformationEx(); test_GetSystemCpuSetInformation(); test_largepages(); + test_large_page_file_mapping(); test_ProcThreadAttributeList(); test_SuspendProcessState(); test_SuspendProcessNewThread();
-
Vibhav Pant -
Vibhav Pant (＠vibhavp)