X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fwin32_capture.c;h=0115ed0d3c6e65ea18e5498499947757d46ef219;hp=2e501dba3bd5b48480fc4ee53bf40f28fcb409e3;hb=b5b9681794d1f5f13350e3567f6f6e74f5c779cf;hpb=8d6441232a9080d8f7d9db839c4c3e17349c4a77 diff --git a/src/win32_capture.c b/src/win32_capture.c index 2e501dba..0115ed0d 100644 --- a/src/win32_capture.c +++ b/src/win32_capture.c @@ -1,9 +1,11 @@ /* * win32_capture.c - Windows-specific code for capturing files into a WIM image. + * + * This now uses the native Windows NT API a lot and not just Win32. */ /* - * Copyright (C) 2013 Eric Biggers + * Copyright (C) 2013, 2014 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. * @@ -30,183 +32,194 @@ #include "wimlib/win32_common.h" #include "wimlib/capture.h" +#include "wimlib/dentry.h" +#include "wimlib/encoding.h" #include "wimlib/endianness.h" #include "wimlib/error.h" #include "wimlib/lookup_table.h" #include "wimlib/paths.h" #include "wimlib/reparse.h" -#define MAX_GET_SD_ACCESS_DENIED_WARNINGS 1 -#define MAX_GET_SACL_PRIV_NOTHELD_WARNINGS 1 -#define MAX_CAPTURE_LONG_PATH_WARNINGS 5 +#include -struct win32_capture_state { +struct winnt_scan_stats { unsigned long num_get_sd_access_denied; unsigned long num_get_sacl_priv_notheld; unsigned long num_long_path_warnings; }; +static inline const wchar_t * +printable_path(const wchar_t *full_path) +{ + /* Skip over \\?\ or \??\ */ + return full_path + 4; +} -static const wchar_t *capture_access_denied_msg = -L" If you are not running this program as the administrator, you may\n" - " need to do so, so that all data and metadata can be backed up.\n" - " Otherwise, there may be no way to access the desired data or\n" - " metadata without taking ownership of the file or directory.\n" - ; +/* + * If cur_dir is not NULL, open an existing file relative to the already-open + * directory cur_dir. + * + * Otherwise, open the file specified by @path, which must be a Windows NT + * namespace path. + */ +static NTSTATUS +winnt_openat(HANDLE cur_dir, const wchar_t *path, size_t path_nchars, + ACCESS_MASK perms, HANDLE *h_ret) +{ + UNICODE_STRING name; + OBJECT_ATTRIBUTES attr; + IO_STATUS_BLOCK iosb; + NTSTATUS status; + + name.Length = path_nchars * sizeof(wchar_t); + name.MaximumLength = name.Length + sizeof(wchar_t); + name.Buffer = (wchar_t *)path; + + attr.Length = sizeof(attr); + attr.RootDirectory = cur_dir; + attr.ObjectName = &name; + attr.Attributes = 0; + attr.SecurityDescriptor = NULL; + attr.SecurityQualityOfService = NULL; + +retry: + status = (*func_NtOpenFile)(h_ret, perms, &attr, &iosb, + FILE_SHARE_VALID_FLAGS, + FILE_OPEN_REPARSE_POINT | + FILE_OPEN_FOR_BACKUP_INTENT | + FILE_SYNCHRONOUS_IO_NONALERT | + FILE_SEQUENTIAL_ONLY); + if (!NT_SUCCESS(status)) { + /* Try requesting fewer permissions */ + if (status == STATUS_ACCESS_DENIED || + status == STATUS_PRIVILEGE_NOT_HELD) { + if (perms & ACCESS_SYSTEM_SECURITY) { + perms &= ~ACCESS_SYSTEM_SECURITY; + goto retry; + } + if (perms & READ_CONTROL) { + perms &= ~READ_CONTROL; + goto retry; + } + } + } + return status; +} +/* Read the first @size bytes from the file, or named data stream of a file, + * from which the stream entry @lte was created. */ int -read_win32_file_prefix(const struct wim_lookup_table_entry *lte, - u64 size, - consume_data_callback_t cb, - void *ctx_or_buf, - int _ignored_flags) +read_winnt_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, + consume_data_callback_t cb, void *cb_ctx) { - int ret = 0; - void *out_buf; - DWORD err; + const wchar_t *path; + HANDLE h; + NTSTATUS status; + u8 buf[BUFFER_SIZE]; u64 bytes_remaining; + int ret; - HANDLE hFile = win32_open_file_data_only(lte->file_on_disk); - if (hFile == INVALID_HANDLE_VALUE) { - err = GetLastError(); - ERROR("Failed to open \"%ls\"", lte->file_on_disk); - win32_error(err); + /* This is an NT namespace path. */ + path = lte->file_on_disk; + + status = winnt_openat(NULL, path, wcslen(path), + FILE_READ_DATA | SYNCHRONIZE, &h); + if (!NT_SUCCESS(status)) { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Can't open for reading " + "(status=0x%08"PRIx32")", + printable_path(path), (u32)status); return WIMLIB_ERR_OPEN; } - if (cb) - out_buf = alloca(WIM_CHUNK_SIZE); - else - out_buf = ctx_or_buf; - + ret = 0; bytes_remaining = size; while (bytes_remaining) { - DWORD bytesToRead, bytesRead; - - bytesToRead = min(WIM_CHUNK_SIZE, bytes_remaining); - if (!ReadFile(hFile, out_buf, bytesToRead, &bytesRead, NULL) || - bytesRead != bytesToRead) - { - err = GetLastError(); - ERROR("Failed to read data from \"%ls\"", lte->file_on_disk); - win32_error(err); + IO_STATUS_BLOCK iosb; + ULONG count; + ULONG bytes_read; + + count = min(sizeof(buf), bytes_remaining); + + status = (*func_NtReadFile)(h, NULL, NULL, NULL, + &iosb, buf, count, NULL, NULL); + if (!NT_SUCCESS(status)) { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Error reading data " + "(status=0x%08"PRIx32")", + printable_path(path), (u32)status); ret = WIMLIB_ERR_READ; break; } - bytes_remaining -= bytesRead; - if (cb) { - ret = (*cb)(out_buf, bytesRead, ctx_or_buf); - if (ret) - break; - } else { - out_buf += bytesRead; - } + + bytes_read = iosb.Information; + + bytes_remaining -= bytes_read; + ret = (*cb)(buf, bytes_read, cb_ctx); + if (ret) + break; } - CloseHandle(hFile); + (*func_NtClose)(h); return ret; } struct win32_encrypted_read_ctx { consume_data_callback_t read_prefix_cb; - void *read_prefix_ctx_or_buf; + void *read_prefix_ctx; int wimlib_err_code; - void *buf; - size_t buf_filled; u64 bytes_remaining; }; static DWORD WINAPI -win32_encrypted_export_cb(unsigned char *_data, void *_ctx, unsigned long len) +win32_encrypted_export_cb(unsigned char *data, void *_ctx, unsigned long len) { - const void *data = _data; struct win32_encrypted_read_ctx *ctx = _ctx; int ret; + size_t bytes_to_consume = min(len, ctx->bytes_remaining); - DEBUG("len = %lu", len); - if (ctx->read_prefix_cb) { - /* The length of the buffer passed to the ReadEncryptedFileRaw() - * export callback is undocumented, so we assume it may be of - * arbitrary size. */ - size_t bytes_to_buffer = min(ctx->bytes_remaining - ctx->buf_filled, - len); - while (bytes_to_buffer) { - size_t bytes_to_copy_to_buf = - min(bytes_to_buffer, WIM_CHUNK_SIZE - ctx->buf_filled); - - memcpy(ctx->buf + ctx->buf_filled, data, - bytes_to_copy_to_buf); - ctx->buf_filled += bytes_to_copy_to_buf; - data += bytes_to_copy_to_buf; - bytes_to_buffer -= bytes_to_copy_to_buf; - - if (ctx->buf_filled == WIM_CHUNK_SIZE || - ctx->buf_filled == ctx->bytes_remaining) - { - ret = (*ctx->read_prefix_cb)(ctx->buf, - ctx->buf_filled, - ctx->read_prefix_ctx_or_buf); - if (ret) { - ctx->wimlib_err_code = ret; - /* Shouldn't matter what error code is returned - * here, as long as it isn't ERROR_SUCCESS. */ - return ERROR_READ_FAULT; - } - ctx->bytes_remaining -= ctx->buf_filled; - ctx->buf_filled = 0; - } - } - } else { - size_t len_to_copy = min(len, ctx->bytes_remaining); - ctx->read_prefix_ctx_or_buf = mempcpy(ctx->read_prefix_ctx_or_buf, - data, - len_to_copy); - ctx->bytes_remaining -= len_to_copy; + if (bytes_to_consume == 0) + return ERROR_SUCCESS; + + ret = (*ctx->read_prefix_cb)(data, bytes_to_consume, ctx->read_prefix_ctx); + if (ret) { + ctx->wimlib_err_code = ret; + /* Shouldn't matter what error code is returned here, as long as + * it isn't ERROR_SUCCESS. */ + return ERROR_READ_FAULT; } + ctx->bytes_remaining -= bytes_to_consume; return ERROR_SUCCESS; } int read_win32_encrypted_file_prefix(const struct wim_lookup_table_entry *lte, u64 size, - consume_data_callback_t cb, - void *ctx_or_buf, - int _ignored_flags) + consume_data_callback_t cb, void *cb_ctx) { struct win32_encrypted_read_ctx export_ctx; DWORD err; void *file_ctx; int ret; - DEBUG("Reading %"PRIu64" bytes from encryted file \"%ls\"", - size, lte->file_on_disk); - export_ctx.read_prefix_cb = cb; - export_ctx.read_prefix_ctx_or_buf = ctx_or_buf; + export_ctx.read_prefix_ctx = cb_ctx; export_ctx.wimlib_err_code = 0; - if (cb) { - export_ctx.buf = MALLOC(WIM_CHUNK_SIZE); - if (!export_ctx.buf) - return WIMLIB_ERR_NOMEM; - } else { - export_ctx.buf = NULL; - } - export_ctx.buf_filled = 0; export_ctx.bytes_remaining = size; - err = OpenEncryptedFileRawW(lte->file_on_disk, 0, &file_ctx); + err = OpenEncryptedFileRaw(lte->file_on_disk, 0, &file_ctx); if (err != ERROR_SUCCESS) { - ERROR("Failed to open encrypted file \"%ls\" for raw read", - lte->file_on_disk); - win32_error(err); - ret = WIMLIB_ERR_OPEN; - goto out_free_buf; + set_errno_from_win32_error(err); + ERROR_WITH_ERRNO("Failed to open encrypted file \"%ls\" " + "for raw read", + printable_path(lte->file_on_disk)); + return WIMLIB_ERR_OPEN; } err = ReadEncryptedFileRaw(win32_encrypted_export_cb, &export_ctx, file_ctx); if (err != ERROR_SUCCESS) { - ERROR("Failed to read encrypted file \"%ls\"", - lte->file_on_disk); - win32_error(err); + set_errno_from_win32_error(err); + ERROR_WITH_ERRNO("Failed to read encrypted file \"%ls\"", + printable_path(lte->file_on_disk)); ret = export_ctx.wimlib_err_code; if (ret == 0) ret = WIMLIB_ERR_READ; @@ -214,351 +227,468 @@ read_win32_encrypted_file_prefix(const struct wim_lookup_table_entry *lte, ERROR("Only could read %"PRIu64" of %"PRIu64" bytes from " "encryted file \"%ls\"", size - export_ctx.bytes_remaining, size, - lte->file_on_disk); + printable_path(lte->file_on_disk)); ret = WIMLIB_ERR_READ; } else { ret = 0; } CloseEncryptedFileRaw(file_ctx); -out_free_buf: - FREE(export_ctx.buf); return ret; } - -static u64 -FILETIME_to_u64(const FILETIME *ft) -{ - return ((u64)ft->dwHighDateTime << 32) | (u64)ft->dwLowDateTime; -} - -static int -win32_get_short_name(struct wim_dentry *dentry, const wchar_t *path) +/* + * Load the short name of a file into a WIM dentry. + */ +static NTSTATUS +winnt_get_short_name(HANDLE h, struct wim_dentry *dentry) { - WIN32_FIND_DATAW dat; - HANDLE hFind; - int ret = 0; - - /* If we can't read the short filename for some reason, we just ignore - * the error and assume the file has no short name. I don't think this - * should be an issue, since the short names are essentially obsolete - * anyway. */ - hFind = FindFirstFileW(path, &dat); - if (hFind != INVALID_HANDLE_VALUE) { - if (dat.cAlternateFileName[0] != L'\0') { - DEBUG("\"%ls\": short name \"%ls\"", path, dat.cAlternateFileName); - size_t short_name_nbytes = wcslen(dat.cAlternateFileName) * - sizeof(wchar_t); - size_t n = short_name_nbytes + sizeof(wchar_t); - dentry->short_name = MALLOC(n); - if (dentry->short_name) { - memcpy(dentry->short_name, dat.cAlternateFileName, n); - dentry->short_name_nbytes = short_name_nbytes; - } else { - ret = WIMLIB_ERR_NOMEM; - } - } - FindClose(hFind); + /* It's not any harder to just make the NtQueryInformationFile() system + * call ourselves, and it saves a dumb call to FindFirstFile() which of + * course has to create its own handle. */ + NTSTATUS status; + IO_STATUS_BLOCK iosb; + u8 buf[128] _aligned_attribute(8); + const FILE_NAME_INFORMATION *info; + + status = (*func_NtQueryInformationFile)(h, &iosb, buf, sizeof(buf), + FileAlternateNameInformation); + info = (const FILE_NAME_INFORMATION *)buf; + if (NT_SUCCESS(status) && info->FileNameLength != 0) { + dentry->short_name = utf16le_dupz(info->FileName, + info->FileNameLength); + if (!dentry->short_name) + return STATUS_NO_MEMORY; + dentry->short_name_nbytes = info->FileNameLength; } - return ret; + return status; } -static int -win32_get_security_descriptor(struct wim_dentry *dentry, +/* + * Load the security descriptor of a file into the corresponding inode, and the + * WIM image's security descriptor set. + */ +static NTSTATUS +winnt_get_security_descriptor(HANDLE h, struct wim_inode *inode, struct wim_sd_set *sd_set, - const wchar_t *path, - struct win32_capture_state *state, - int add_flags) + struct winnt_scan_stats *stats, int add_flags) { SECURITY_INFORMATION requestedInformation; - DWORD lenNeeded = 0; - BOOL status; - DWORD err; - unsigned long n; + u8 _buf[4096] _aligned_attribute(8); + u8 *buf; + ULONG bufsize; + ULONG len_needed; + NTSTATUS status; requestedInformation = DACL_SECURITY_INFORMATION | SACL_SECURITY_INFORMATION | OWNER_SECURITY_INFORMATION | GROUP_SECURITY_INFORMATION; -again: - /* Request length of security descriptor */ - status = GetFileSecurityW(path, requestedInformation, - NULL, 0, &lenNeeded); - err = GetLastError(); - if (!status && err == ERROR_INSUFFICIENT_BUFFER) { - DWORD len = lenNeeded; - char buf[len]; - if (GetFileSecurityW(path, requestedInformation, - (PSECURITY_DESCRIPTOR)buf, len, &lenNeeded)) - { - int security_id = sd_set_add_sd(sd_set, buf, len); - if (security_id < 0) - return WIMLIB_ERR_NOMEM; - else { - dentry->d_inode->i_security_id = security_id; - return 0; + buf = _buf; + bufsize = sizeof(_buf); + + /* + * We need the file's security descriptor in + * SECURITY_DESCRIPTOR_RELATIVE format, and we currently have a handle + * opened with as many relevant permissions as possible. At this point, + * on Windows there are a number of options for reading a file's + * security descriptor: + * + * GetFileSecurity(): This takes in a path and returns the + * SECURITY_DESCRIPTOR_RELATIVE. Problem: this uses an internal handle, + * not ours, and the handle created internally doesn't specify + * FILE_FLAG_BACKUP_SEMANTICS. Therefore there can be access denied + * errors on some files and directories, even when running as the + * Administrator. + * + * GetSecurityInfo(): This takes in a handle and returns the security + * descriptor split into a bunch of different parts. This should work, + * but it's dumb because we have to put the security descriptor back + * together again. + * + * BackupRead(): This can read the security descriptor, but this is a + * difficult-to-use API, probably only works as the Administrator, and + * the format of the returned data is not well documented. + * + * NtQuerySecurityObject(): This is exactly what we need, as it takes + * in a handle and returns the security descriptor in + * SECURITY_DESCRIPTOR_RELATIVE format. Only problem is that it's a + * ntdll function and therefore not officially part of the Win32 API. + * Oh well. + */ + while (!(NT_SUCCESS(status = (*func_NtQuerySecurityObject)(h, + requestedInformation, + (PSECURITY_DESCRIPTOR)buf, + bufsize, + &len_needed)))) + { + switch (status) { + case STATUS_BUFFER_TOO_SMALL: + wimlib_assert(buf == _buf); + buf = MALLOC(len_needed); + if (!buf) + return STATUS_NO_MEMORY; + bufsize = len_needed; + break; + case STATUS_PRIVILEGE_NOT_HELD: + case STATUS_ACCESS_DENIED: + if (add_flags & WIMLIB_ADD_FLAG_STRICT_ACLS) { + default: + /* Permission denied in STRICT_ACLS mode, or + * unknown error. */ + goto out_free_buf; } - } else { - err = GetLastError(); - } - } - - if (add_flags & WIMLIB_ADD_FLAG_STRICT_ACLS) - goto fail; - - switch (err) { - case ERROR_PRIVILEGE_NOT_HELD: - if (requestedInformation & SACL_SECURITY_INFORMATION) { - n = state->num_get_sacl_priv_notheld++; - requestedInformation &= ~SACL_SECURITY_INFORMATION; - if (n < MAX_GET_SACL_PRIV_NOTHELD_WARNINGS) { - WARNING( -"We don't have enough privileges to read the full security\n" -" descriptor of \"%ls\"!\n" -" Re-trying with SACL omitted.\n", path); - } else if (n == MAX_GET_SACL_PRIV_NOTHELD_WARNINGS) { - WARNING( -"Suppressing further privileges not held error messages when reading\n" -" security descriptors."); + if (requestedInformation & SACL_SECURITY_INFORMATION) { + /* Try again without the SACL. */ + stats->num_get_sacl_priv_notheld++; + requestedInformation &= ~SACL_SECURITY_INFORMATION; + break; } - goto again; - } - /* Fall through */ - case ERROR_ACCESS_DENIED: - n = state->num_get_sd_access_denied++; - if (n < MAX_GET_SD_ACCESS_DENIED_WARNINGS) { - WARNING("Failed to read security descriptor of \"%ls\": " - "Access denied!\n%ls", path, capture_access_denied_msg); - } else if (n == MAX_GET_SD_ACCESS_DENIED_WARNINGS) { - WARNING("Suppressing further access denied errors messages i" - "when reading security descriptors"); + /* Fake success (useful when capturing as + * non-Administrator). */ + stats->num_get_sd_access_denied++; + status = STATUS_SUCCESS; + goto out_free_buf; } - return 0; - default: -fail: - ERROR("Failed to read security descriptor of \"%ls\"", path); - win32_error(err); - return WIMLIB_ERR_READ; } + + /* Add the security descriptor to the WIM image, and save its ID in + * file's inode. */ + inode->i_security_id = sd_set_add_sd(sd_set, buf, len_needed); + if (unlikely(inode->i_security_id < 0)) + status = STATUS_NO_MEMORY; +out_free_buf: + if (unlikely(buf != _buf)) + FREE(buf); + return status; } static int -win32_build_dentry_tree_recursive(struct wim_dentry **root_ret, - wchar_t *path, - size_t path_num_chars, +winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret, + HANDLE cur_dir, + wchar_t *full_path, + size_t full_path_nchars, + const wchar_t *filename, + size_t filename_nchars, struct add_image_params *params, - struct win32_capture_state *state, - unsigned vol_flags); + struct winnt_scan_stats *stats, + u32 vol_flags); -/* Reads the directory entries of directory using a Win32 API and recursively - * calls win32_build_dentry_tree() on them. */ static int -win32_recurse_directory(struct wim_dentry *root, - wchar_t *dir_path, - size_t dir_path_num_chars, +winnt_recurse_directory(HANDLE h, + wchar_t *full_path, + size_t full_path_nchars, + struct wim_dentry *parent, struct add_image_params *params, - struct win32_capture_state *state, - unsigned vol_flags) + struct winnt_scan_stats *stats, + u32 vol_flags) { - WIN32_FIND_DATAW dat; - HANDLE hFind; - DWORD err; + void *buf; + const size_t bufsize = 8192; + IO_STATUS_BLOCK iosb; + NTSTATUS status; int ret; - DEBUG("Recurse to directory \"%ls\"", dir_path); - - /* Begin reading the directory by calling FindFirstFileW. Unlike UNIX - * opendir(), FindFirstFileW has file globbing built into it. But this - * isn't what we actually want, so just add a dummy glob to get all - * entries. */ - dir_path[dir_path_num_chars] = OS_PREFERRED_PATH_SEPARATOR; - dir_path[dir_path_num_chars + 1] = L'*'; - dir_path[dir_path_num_chars + 2] = L'\0'; - hFind = FindFirstFileW(dir_path, &dat); - dir_path[dir_path_num_chars] = L'\0'; - - if (hFind == INVALID_HANDLE_VALUE) { - err = GetLastError(); - if (err == ERROR_FILE_NOT_FOUND) { - return 0; - } else { - ERROR("Failed to read directory \"%ls\"", dir_path); - win32_error(err); - return WIMLIB_ERR_READ; - } - } - ret = 0; - do { - /* Skip . and .. entries */ - if (dat.cFileName[0] == L'.' && - (dat.cFileName[1] == L'\0' || - (dat.cFileName[1] == L'.' && - dat.cFileName[2] == L'\0'))) - continue; - size_t filename_len = wcslen(dat.cFileName); - - dir_path[dir_path_num_chars] = OS_PREFERRED_PATH_SEPARATOR; - wmemcpy(dir_path + dir_path_num_chars + 1, - dat.cFileName, - filename_len + 1); - - struct wim_dentry *child; - size_t path_len = dir_path_num_chars + 1 + filename_len; - ret = win32_build_dentry_tree_recursive(&child, - dir_path, - path_len, + buf = MALLOC(bufsize); + if (!buf) + return WIMLIB_ERR_NOMEM; + + /* Using NtQueryDirectoryFile() we can re-use the same open handle, + * which we opened with FILE_FLAG_BACKUP_SEMANTICS. */ + + while (NT_SUCCESS(status = (*func_NtQueryDirectoryFile)(h, NULL, NULL, NULL, + &iosb, buf, bufsize, + FileNamesInformation, + FALSE, NULL, FALSE))) + { + const FILE_NAMES_INFORMATION *info = buf; + for (;;) { + if (!(info->FileNameLength == 2 && info->FileName[0] == L'.') && + !(info->FileNameLength == 4 && info->FileName[0] == L'.' && + info->FileName[1] == L'.')) + { + wchar_t *p; + struct wim_dentry *child; + + p = full_path + full_path_nchars; + *p++ = L'\\'; + p = wmempcpy(p, info->FileName, + info->FileNameLength / 2); + *p = '\0'; + + ret = winnt_build_dentry_tree_recursive( + &child, + h, + full_path, + p - full_path, + full_path + full_path_nchars + 1, + info->FileNameLength / 2, params, - state, + stats, vol_flags); - dir_path[dir_path_num_chars] = L'\0'; - if (ret) - goto out_find_close; - if (child) - dentry_add_child(root, child); - } while (FindNextFileW(hFind, &dat)); - err = GetLastError(); - if (err != ERROR_NO_MORE_FILES) { - ERROR("Failed to read directory \"%ls\"", dir_path); - win32_error(err); - if (ret == 0) - ret = WIMLIB_ERR_READ; + + full_path[full_path_nchars] = L'\0'; + + if (ret) + goto out_free_buf; + if (child) + dentry_add_child(parent, child); + } + if (info->NextEntryOffset == 0) + break; + info = (const FILE_NAMES_INFORMATION *) + ((const u8 *)info + info->NextEntryOffset); + } + } + + if (unlikely(status != STATUS_NO_MORE_FILES)) { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Can't read directory " + "(status=0x%08"PRIx32")", + printable_path(full_path), (u32)status); + ret = WIMLIB_ERR_READ; } -out_find_close: - FindClose(hFind); +out_free_buf: + FREE(buf); return ret; } -/* Reparse point fixup status code */ +/* Reparse point fixup status code */ enum rp_status { - /* Reparse point corresponded to an absolute symbolic link or junction - * point that pointed outside the directory tree being captured, and - * therefore was excluded. */ - RP_EXCLUDED = 0x0, - - /* Reparse point was not fixed as it was either a relative symbolic - * link, a mount point, or something else we could not understand. */ - RP_NOT_FIXED = 0x1, - - /* Reparse point corresponded to an absolute symbolic link or junction - * point that pointed inside the directory tree being captured, where - * the target was specified by a "full" \??\ prefixed path, and - * therefore was fixed to be relative to the root of the directory tree - * being captured. */ - RP_FIXED_FULLPATH = 0x2, - - /* Same as RP_FIXED_FULLPATH, except the absolute link target did not - * have the \??\ prefix. It may have begun with a drive letter though. - * */ - RP_FIXED_ABSPATH = 0x4, - - /* Either RP_FIXED_FULLPATH or RP_FIXED_ABSPATH. */ - RP_FIXED = RP_FIXED_FULLPATH | RP_FIXED_ABSPATH, + /* Reparse point should be excluded from capture */ + RP_EXCLUDED = -0, + + /* Reparse point will be captured literally (no fixup) */ + RP_NOT_FIXED = -1, + + /* Reparse point will be captured with fixup */ + RP_FIXED = -2, }; -/* Given the "substitute name" target of a Windows reparse point, try doing a - * fixup where we change it to be absolute relative to the root of the directory - * tree being captured. +static bool +file_has_ino_and_dev(HANDLE h, u64 ino, u64 dev) +{ + NTSTATUS status; + IO_STATUS_BLOCK iosb; + FILE_INTERNAL_INFORMATION int_info; + FILE_FS_VOLUME_INFORMATION vol_info; + + status = (*func_NtQueryInformationFile)(h, &iosb, + &int_info, sizeof(int_info), + FileInternalInformation); + if (!NT_SUCCESS(status)) + return false; + + if (int_info.IndexNumber.QuadPart != ino) + return false; + + status = (*func_NtQueryVolumeInformationFile)(h, &iosb, + &vol_info, sizeof(vol_info), + FileFsVolumeInformation); + if (!(NT_SUCCESS(status) || status == STATUS_BUFFER_OVERFLOW)) + return false; + + if (iosb.Information < + offsetof(FILE_FS_VOLUME_INFORMATION, VolumeSerialNumber) + + sizeof(vol_info.VolumeSerialNumber)) + return false; + + return (vol_info.VolumeSerialNumber == dev); +} + +/* + * Given an (expected) NT namespace symbolic link or junction target @target of + * length @target_nbytes, determine if a prefix of the target points to a file + * identified by @capture_root_ino and @capture_root_dev. + * + * If yes, return a pointer to the portion of the link following this prefix. * - * Note that this is only executed when WIMLIB_ADD_FLAG_RPFIX has been - * set. + * If no, return NULL. * - * @capture_root_ino and @capture_root_dev indicate the inode number and device - * of the root of the directory tree being captured. They are meant to identify - * this directory (as an alternative to its actual path, which could potentially - * be reached via multiple destinations due to other symbolic links). This may - * not work properly on FAT, which doesn't seem to supply proper inode numbers - * or file IDs. However, FAT doesn't support reparse points so this function - * wouldn't even be called anyway. + * If the link target does not appear to be a valid NT namespace path, return + * @target itself. */ -static enum rp_status -win32_capture_maybe_rpfix_target(wchar_t *target, u16 *target_nbytes_p, - u64 capture_root_ino, u64 capture_root_dev, - u32 rptag) +static const wchar_t * +winnt_get_root_relative_target(const wchar_t *target, size_t target_nbytes, + u64 capture_root_ino, u64 capture_root_dev) { - u16 target_nchars = *target_nbytes_p / 2; - size_t stripped_chars; - wchar_t *orig_target; - int ret; + UNICODE_STRING name; + OBJECT_ATTRIBUTES attr; + IO_STATUS_BLOCK iosb; + NTSTATUS status; + const wchar_t *target_end; + const wchar_t *p; + + target_end = target + (target_nbytes / sizeof(wchar_t)); + + /* Empty path??? */ + if (target_end == target) + return target; + + /* No leading slash??? */ + if (target[0] != L'\\') + return target; + + /* UNC path??? */ + if ((target_end - target) >= 2 && + target[0] == L'\\' && target[1] == L'\\') + return target; + + attr.Length = sizeof(attr); + attr.RootDirectory = NULL; + attr.ObjectName = &name; + attr.Attributes = 0; + attr.SecurityDescriptor = NULL; + attr.SecurityQualityOfService = NULL; + + name.Buffer = (wchar_t *)target; + name.Length = 0; + p = target; + do { + HANDLE h; + const wchar_t *orig_p = p; + + /* Skip non-backslashes */ + while (p != target_end && *p != L'\\') + p++; + + /* Skip backslashes */ + while (p != target_end && *p == L'\\') + p++; + + /* Append path component */ + name.Length += (p - orig_p) * sizeof(wchar_t); + name.MaximumLength = name.Length; + + /* Try opening the file */ + status = (*func_NtOpenFile) (&h, + FILE_READ_ATTRIBUTES | FILE_TRAVERSE, + &attr, + &iosb, + FILE_SHARE_VALID_FLAGS, + FILE_OPEN_FOR_BACKUP_INTENT); + + if (NT_SUCCESS(status)) { + /* Reset root directory */ + if (attr.RootDirectory) + (*func_NtClose)(attr.RootDirectory); + attr.RootDirectory = h; + name.Buffer = (wchar_t *)p; + name.Length = 0; + + if (file_has_ino_and_dev(h, capture_root_ino, + capture_root_dev)) + goto out_close_root_dir; + } + } while (p != target_end); - ret = parse_substitute_name(target, *target_nbytes_p, rptag); - if (ret < 0) - return RP_NOT_FIXED; - stripped_chars = ret; - if (stripped_chars) - stripped_chars -= 2; - target[target_nchars] = L'\0'; - orig_target = target; - target = capture_fixup_absolute_symlink(target + stripped_chars, - capture_root_ino, capture_root_dev); - if (!target) - return RP_EXCLUDED; - target_nchars = wcslen(target); - wmemmove(orig_target + stripped_chars, target, target_nchars + 1); - *target_nbytes_p = (target_nchars + stripped_chars) * sizeof(wchar_t); - DEBUG("Fixed reparse point (new target: \"%ls\")", orig_target); - if (stripped_chars) - return RP_FIXED_FULLPATH; - else - return RP_FIXED_ABSPATH; + p = NULL; + +out_close_root_dir: + if (attr.RootDirectory) + (*func_NtClose)(attr.RootDirectory); + return p; } -/* Returns: `enum rp_status' value on success; negative WIMLIB_ERR_* value on - * failure. */ static int -win32_capture_try_rpfix(u8 *rpbuf, u16 *rpbuflen_p, - u64 capture_root_ino, u64 capture_root_dev, - const wchar_t *path) +winnt_try_rpfix(u8 *rpbuf, u16 *rpbuflen_p, + u64 capture_root_ino, u64 capture_root_dev, + const wchar_t *path, struct add_image_params *params) { struct reparse_data rpdata; - int ret; - enum rp_status rp_status; + const wchar_t *rel_target; - ret = parse_reparse_data(rpbuf, *rpbuflen_p, &rpdata); - if (ret) - return -ret; - - rp_status = win32_capture_maybe_rpfix_target(rpdata.substitute_name, - &rpdata.substitute_name_nbytes, - capture_root_ino, - capture_root_dev, - le32_to_cpu(*(le32*)rpbuf)); - if (rp_status & RP_FIXED) { - wimlib_assert(rpdata.substitute_name_nbytes % 2 == 0); - utf16lechar substitute_name_copy[rpdata.substitute_name_nbytes / 2]; - wmemcpy(substitute_name_copy, rpdata.substitute_name, - rpdata.substitute_name_nbytes / 2); - rpdata.substitute_name = substitute_name_copy; - rpdata.print_name = substitute_name_copy; - rpdata.print_name_nbytes = rpdata.substitute_name_nbytes; - if (rp_status == RP_FIXED_FULLPATH) { - /* "full path", meaning \??\ prefixed. We should not - * include this prefix in the print name, as it is - * apparently meant for the filesystem driver only. */ - rpdata.print_name += 4; - rpdata.print_name_nbytes -= 8; - } - ret = make_reparse_buffer(&rpdata, rpbuf, rpbuflen_p); - if (ret == 0) - ret = rp_status; - else - ret = -ret; - } else { - if (rp_status == RP_EXCLUDED) { - size_t print_name_nchars = rpdata.print_name_nbytes / 2; - wchar_t print_name0[print_name_nchars + 1]; - print_name0[print_name_nchars] = L'\0'; - wmemcpy(print_name0, rpdata.print_name, print_name_nchars); - WARNING("Ignoring %ls pointing out of capture directory:\n" - " \"%ls\" -> \"%ls\"\n" - " (Use --norpfix to capture all symbolic links " - "and junction points as-is)", - (rpdata.rptag == WIM_IO_REPARSE_TAG_SYMLINK) ? - L"absolute symbolic link" : L"junction point", - path, print_name0); - } - ret = rp_status; + if (parse_reparse_data(rpbuf, *rpbuflen_p, &rpdata)) { + /* Couldn't even understand the reparse data. Don't try the + * fixup. */ + return RP_NOT_FIXED; } - return ret; + + /* + * Don't do reparse point fixups on relative symbolic links. + * + * On Windows, a relative symbolic link is supposed to be identifiable + * by having reparse tag WIM_IO_REPARSE_TAG_SYMLINK and flags + * SYMBOLIC_LINK_RELATIVE. We will use this information, although this + * may not always do what the user expects, since drive-relative + * symbolic links such as "\Users\Public" have SYMBOLIC_LINK_RELATIVE + * set, in addition to truely relative symbolic links such as "Users" or + * "Users\Public". However, WIMGAPI (as of Windows 8.1) has this same + * behavior. + * + * Otherwise, as far as I can tell, the targets of symbolic links that + * are NOT relative, as well as junctions (note: a mountpoint is the + * sames thing as a junction), must be NT namespace paths, for example: + * + * - \??\e:\Users\Public + * - \DosDevices\e:\Users\Public + * - \Device\HardDiskVolume4\Users\Public + * - \??\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public + * - \DosDevices\Volume{c47cb07c-946e-4155-b8f7-052e9cec7628}\Users\Public + */ + if (rpdata.rptag == WIM_IO_REPARSE_TAG_SYMLINK && + (rpdata.rpflags & SYMBOLIC_LINK_RELATIVE)) + return RP_NOT_FIXED; + + rel_target = winnt_get_root_relative_target(rpdata.substitute_name, + rpdata.substitute_name_nbytes, + capture_root_ino, + capture_root_dev); + if (!rel_target) { + /* Target points outside of the tree being captured. Exclude + * this reparse point from the capture (but inform the library + * user). */ + size_t print_name_nchars = rpdata.print_name_nbytes / sizeof(wchar_t); + wchar_t print_name0[print_name_nchars + 1]; + print_name0[print_name_nchars] = L'\0'; + wmemcpy(print_name0, rpdata.print_name, print_name_nchars); + + params->progress.scan.cur_path = printable_path(path); + params->progress.scan.symlink_target = print_name0; + int ret = do_capture_progress(params, + WIMLIB_SCAN_DENTRY_EXCLUDED_SYMLINK, + NULL); + if (ret) + return ret; + return RP_EXCLUDED; + } + + if (rel_target == rpdata.substitute_name) { + /* Weird target --- keep the reparse point and don't mess with + * it. */ + return RP_NOT_FIXED; + } + + /* We have an absolute target pointing within the directory being + * captured, @rel_target is the suffix of the link target that is the + * part relative to the directory being captured. + * + * We will cut off the prefix before this part (which is the path to the + * directory being captured) and add a dummy prefix. Since the process + * will need to be reversed when applying the image, it shouldn't matter + * what exactly the prefix is, as long as it looks like an absolute + * path. + */ + + { + size_t rel_target_nbytes = + rpdata.substitute_name_nbytes - ((const u8 *)rel_target - + (const u8 *)rpdata.substitute_name); + size_t rel_target_nchars = rel_target_nbytes / sizeof(wchar_t); + + wchar_t tmp[rel_target_nchars + 7]; + + wmemcpy(tmp, L"\\??\\X:\\", 7); + wmemcpy(tmp + 7, rel_target, rel_target_nchars); + + rpdata.substitute_name = tmp; + rpdata.substitute_name_nbytes = rel_target_nbytes + (7 * sizeof(wchar_t)); + rpdata.print_name = tmp + 4; + rpdata.print_name_nbytes = rel_target_nbytes + (3 * sizeof(wchar_t)); + + if (make_reparse_buffer(&rpdata, rpbuf, rpbuflen_p)) + return RP_NOT_FIXED; + } + return RP_FIXED; } /* @@ -566,77 +696,67 @@ win32_capture_try_rpfix(u8 *rpbuf, u16 *rpbuflen_p, * fixing the targets of absolute symbolic links and junction points to be * relative to the root of capture. * - * @hFile: Open handle to the reparse point. - * @path: Path to the reparse point. Used for error messages only. - * @params: Additional parameters, including whether to do reparse point fixups - * or not. - * @rpbuf: Buffer of length at least REPARSE_POINT_MAX_SIZE bytes into which - * the reparse point buffer will be loaded. - * @rpbuflen_ret: On success, the length of the reparse point buffer in bytes - * is written to this location. - * - * Returns: - * On success, returns an `enum rp_status' value that indicates if and/or - * how the reparse point fixup was done. + * @h: + * Open handle to the reparse point file. + * @path: + * Path to the reparse point file. + * @params: + * Capture parameters. add_flags, capture_root_ino, capture_root_dev, + * progfunc, progctx, and progress are used. + * @rpbuf: + * Buffer of length at least REPARSE_POINT_MAX_SIZE bytes into which the + * reparse point buffer will be loaded. + * @rpbuflen_ret: + * On success, the length of the reparse point buffer in bytes is written + * to this location. * - * On failure, returns a negative value that is a negated WIMLIB_ERR_* - * code. + * On success, returns a nonpositive `enum rp_status' value. + * On failure, returns a positive error code. */ static int -win32_get_reparse_data(HANDLE hFile, const wchar_t *path, +winnt_get_reparse_data(HANDLE h, const wchar_t *path, struct add_image_params *params, u8 *rpbuf, u16 *rpbuflen_ret) { - DWORD bytesReturned; + DWORD bytes_returned; u32 reparse_tag; int ret; u16 rpbuflen; - DEBUG("Loading reparse data from \"%ls\"", path); - if (!DeviceIoControl(hFile, FSCTL_GET_REPARSE_POINT, - NULL, /* "Not used with this operation; set to NULL" */ - 0, /* "Not used with this operation; set to 0" */ - rpbuf, /* "A pointer to a buffer that - receives the reparse point data */ - REPARSE_POINT_MAX_SIZE, /* "The size of the output - buffer, in bytes */ - &bytesReturned, - NULL)) + if (!DeviceIoControl(h, FSCTL_GET_REPARSE_POINT, + NULL, 0, rpbuf, REPARSE_POINT_MAX_SIZE, + &bytes_returned, NULL)) { - DWORD err = GetLastError(); - ERROR("Failed to get reparse data of \"%ls\"", path); - win32_error(err); - return -WIMLIB_ERR_READ; + set_errno_from_GetLastError(); + return WIMLIB_ERR_READ; } - if (bytesReturned < 8 || bytesReturned > REPARSE_POINT_MAX_SIZE) { - ERROR("Reparse data on \"%ls\" is invalid", path); - return -WIMLIB_ERR_INVALID_REPARSE_DATA; + + if (unlikely(bytes_returned < 8)) { + errno = EINVAL; + return WIMLIB_ERR_INVALID_REPARSE_DATA; } - rpbuflen = bytesReturned; + rpbuflen = bytes_returned; reparse_tag = le32_to_cpu(*(le32*)rpbuf); + ret = RP_NOT_FIXED; if (params->add_flags & WIMLIB_ADD_FLAG_RPFIX && (reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK || reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT)) { - /* Try doing reparse point fixup */ - ret = win32_capture_try_rpfix(rpbuf, - &rpbuflen, - params->capture_root_ino, - params->capture_root_dev, - path); - } else { - ret = RP_NOT_FIXED; + ret = winnt_try_rpfix(rpbuf, &rpbuflen, + params->capture_root_ino, + params->capture_root_dev, + path, params); } *rpbuflen_ret = rpbuflen; return ret; } static DWORD WINAPI -win32_tally_encrypted_size_cb(unsigned char *_data, void *_ctx, +win32_tally_encrypted_size_cb(unsigned char *_data, void *_size_ret, unsigned long len) { - *(u64*)_ctx += len; + *(u64*)_size_ret += len; return ERROR_SUCCESS; } @@ -647,18 +767,20 @@ win32_get_encrypted_file_size(const wchar_t *path, u64 *size_ret) void *file_ctx; int ret; - *size_ret = 0; - err = OpenEncryptedFileRawW(path, 0, &file_ctx); + err = OpenEncryptedFileRaw(path, 0, &file_ctx); if (err != ERROR_SUCCESS) { - ERROR("Failed to open encrypted file \"%ls\" for raw read", path); - win32_error(err); + set_errno_from_win32_error(err); + ERROR_WITH_ERRNO("Failed to open encrypted file \"%ls\" " + "for raw read", printable_path(path)); return WIMLIB_ERR_OPEN; } + *size_ret = 0; err = ReadEncryptedFileRaw(win32_tally_encrypted_size_cb, size_ret, file_ctx); if (err != ERROR_SUCCESS) { - ERROR("Failed to read raw encrypted data from \"%ls\"", path); - win32_error(err); + set_errno_from_win32_error(err); + ERROR_WITH_ERRNO("Failed to read raw encrypted data from " + "\"%ls\"", printable_path(path)); ret = WIMLIB_ERR_READ; } else { ret = 0; @@ -667,346 +789,422 @@ win32_get_encrypted_file_size(const wchar_t *path, u64 *size_ret) return ret; } -/* Scans an unnamed or named stream of a Win32 file (not a reparse point - * stream); calculates its SHA1 message digest and either creates a `struct - * wim_lookup_table_entry' in memory for it, or uses an existing 'struct - * wim_lookup_table_entry' for an identical stream. - * - * @path: Path to the file (UTF-16LE). - * - * @path_num_chars: Number of 2-byte characters in @path. - * - * @inode: WIM inode to save the stream into. - * - * @lookup_table: Stream lookup table for the WIM. - * - * @dat: A `WIN32_FIND_STREAM_DATA' structure that specifies the - * stream name. - * - * Returns 0 on success; nonzero on failure. - */ -static int -win32_capture_stream(const wchar_t *path, - size_t path_num_chars, - struct wim_inode *inode, - struct wim_lookup_table *lookup_table, - WIN32_FIND_STREAM_DATA *dat) +static bool +get_data_stream_name(const wchar_t *raw_stream_name, size_t raw_stream_name_nchars, + const wchar_t **stream_name_ret, size_t *stream_name_nchars_ret) { - struct wim_ads_entry *ads_entry; - struct wim_lookup_table_entry *lte; - int ret; - wchar_t *stream_name, *colon; - size_t stream_name_nchars; - bool is_named_stream; - wchar_t *spath; - size_t spath_nchars; - size_t spath_buf_nbytes; - const wchar_t *relpath_prefix; - const wchar_t *colonchar; - - DEBUG("Capture \"%ls\" stream \"%ls\"", path, dat->cStreamName); - - /* The stream name should be returned as :NAME:TYPE */ - stream_name = dat->cStreamName; - if (*stream_name != L':') - goto out_invalid_stream_name; - stream_name += 1; - colon = wcschr(stream_name, L':'); - if (colon == NULL) - goto out_invalid_stream_name; - - if (wcscmp(colon + 1, L"$DATA")) { - /* Not a DATA stream */ - ret = 0; - goto out; - } + const wchar_t *sep, *type, *end; - *colon = '\0'; + /* The stream name should be returned as :NAME:TYPE */ + if (raw_stream_name_nchars < 1) + return false; + if (raw_stream_name[0] != L':') + return false; - stream_name_nchars = colon - stream_name; - is_named_stream = (stream_name_nchars != 0); + raw_stream_name++; + raw_stream_name_nchars--; - if (is_named_stream) { - /* Allocate an ADS entry for the named stream. */ - ads_entry = inode_add_ads_utf16le(inode, stream_name, - stream_name_nchars * sizeof(wchar_t)); - if (!ads_entry) { - ret = WIMLIB_ERR_NOMEM; - goto out; + end = raw_stream_name + raw_stream_name_nchars; + + sep = wmemchr(raw_stream_name, L':', raw_stream_name_nchars); + if (!sep) + return false; + + type = sep + 1; + if (end - type != 5) + return false; + + if (wmemcmp(type, L"$DATA", 5)) + return false; + + *stream_name_ret = raw_stream_name; + *stream_name_nchars_ret = sep - raw_stream_name; + return true; +} + +static wchar_t * +build_stream_path(const wchar_t *path, size_t path_nchars, + const wchar_t *stream_name, size_t stream_name_nchars) +{ + size_t stream_path_nchars; + wchar_t *stream_path; + wchar_t *p; + + stream_path_nchars = path_nchars; + if (stream_name_nchars) + stream_path_nchars += 1 + stream_name_nchars; + + stream_path = MALLOC((stream_path_nchars + 1) * sizeof(wchar_t)); + if (stream_path) { + p = wmempcpy(stream_path, path, path_nchars); + if (stream_name_nchars) { + *p++ = L':'; + p = wmempcpy(p, stream_name, stream_name_nchars); } + *p++ = L'\0'; } + return stream_path; +} - /* If zero length stream, no lookup table entry needed. */ - if ((u64)dat->StreamSize.QuadPart == 0) { - ret = 0; - goto out; - } +static int +winnt_scan_stream(const wchar_t *path, size_t path_nchars, + const wchar_t *raw_stream_name, size_t raw_stream_name_nchars, + u64 stream_size, + struct wim_inode *inode, struct list_head *unhashed_streams) +{ + const wchar_t *stream_name; + size_t stream_name_nchars; + struct wim_ads_entry *ads_entry; + wchar_t *stream_path; + struct wim_lookup_table_entry *lte; + u32 stream_id; - /* Create a UTF-16LE string @spath that gives the filename, then a - * colon, then the stream name. Or, if it's an unnamed stream, just the - * filename. It is MALLOC()'ed so that it can be saved in the - * wim_lookup_table_entry if needed. - * - * As yet another special case, relative paths need to be changed to - * begin with an explicit "./" so that, for example, a file t:ads, where - * :ads is the part we added, is not interpreted as a file on the t: - * drive. */ - spath_nchars = path_num_chars; - relpath_prefix = L""; - colonchar = L""; - if (is_named_stream) { - spath_nchars += 1 + stream_name_nchars; - colonchar = L":"; - if (path_num_chars == 1 && !is_any_path_separator(path[0])) { - spath_nchars += 2; - static const wchar_t _relpath_prefix[] = - {L'.', OS_PREFERRED_PATH_SEPARATOR, L'\0'}; - relpath_prefix = _relpath_prefix; - } + /* Given the raw stream name (which is something like + * :streamname:$DATA), extract just the stream name part. + * Ignore any non-$DATA streams. */ + if (!get_data_stream_name(raw_stream_name, raw_stream_name_nchars, + &stream_name, &stream_name_nchars)) + return 0; + + /* If this is a named stream, allocate an ADS entry for it. */ + if (stream_name_nchars) { + ads_entry = inode_add_ads_utf16le(inode, stream_name, + stream_name_nchars * + sizeof(wchar_t)); + if (!ads_entry) + return WIMLIB_ERR_NOMEM; + } else { + ads_entry = NULL; } - spath_buf_nbytes = (spath_nchars + 1) * sizeof(wchar_t); - spath = MALLOC(spath_buf_nbytes); + /* If the stream is empty, no lookup table entry is needed. */ + if (stream_size == 0) + return 0; - swprintf(spath, L"%ls%ls%ls%ls", - relpath_prefix, path, colonchar, stream_name); + /* Build the path to the stream. For unnamed streams, this is simply + * the path to the file. For named streams, this is the path to the + * file, followed by a colon, followed by the stream name. */ + stream_path = build_stream_path(path, path_nchars, + stream_name, stream_name_nchars); + if (!stream_path) + return WIMLIB_ERR_NOMEM; - /* Make a new wim_lookup_table_entry */ + /* Set up the lookup table entry for the stream. */ lte = new_lookup_table_entry(); if (!lte) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_spath; + FREE(stream_path); + return WIMLIB_ERR_NOMEM; } - lte->file_on_disk = spath; - spath = NULL; - if (inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED && !is_named_stream) { + lte->file_on_disk = stream_path; + lte->resource_location = RESOURCE_IN_WINNT_FILE_ON_DISK; + lte->size = stream_size; + if ((inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED) && !ads_entry) { + /* Special case for encrypted file. */ + + /* OpenEncryptedFileRaw() expects Win32 name, not NT name. + * Change \??\ into \\?\ */ + lte->file_on_disk[1] = L'\\'; + wimlib_assert(!wmemcmp(lte->file_on_disk, L"\\\\?\\", 4)); + u64 encrypted_size; + int ret; + + ret = win32_get_encrypted_file_size(lte->file_on_disk, + &encrypted_size); + if (ret) { + free_lookup_table_entry(lte); + return ret; + } + lte->size = encrypted_size; lte->resource_location = RESOURCE_WIN32_ENCRYPTED; - ret = win32_get_encrypted_file_size(path, &encrypted_size); - if (ret) - goto out_free_spath; - lte->resource_entry.original_size = encrypted_size; - } else { - lte->resource_location = RESOURCE_WIN32; - lte->resource_entry.original_size = (u64)dat->StreamSize.QuadPart; } - u32 stream_id; - if (is_named_stream) { + if (ads_entry) { stream_id = ads_entry->stream_id; ads_entry->lte = lte; } else { stream_id = 0; inode->i_lte = lte; } - lookup_table_insert_unhashed(lookup_table, lte, inode, stream_id); - ret = 0; -out_free_spath: - FREE(spath); -out: - return ret; -out_invalid_stream_name: - ERROR("Invalid stream name: \"%ls:%ls\"", path, dat->cStreamName); - ret = WIMLIB_ERR_READ; - goto out; + add_unhashed_stream(lte, inode, stream_id, unhashed_streams); + return 0; } -/* Scans a Win32 file for unnamed and named data streams (not reparse point - * streams). - * - * @path: Path to the file (UTF-16LE). - * - * @path_num_chars: Number of 2-byte characters in @path. - * - * @inode: WIM inode to save the stream into. - * - * @lookup_table: Stream lookup table for the WIM. - * - * @file_size: Size of unnamed data stream. (Used only if alternate - * data streams API appears to be unavailable.) +/* + * Load information about the streams of an open file into a WIM inode. * - * @vol_flags: Flags that specify features of the volume being - * captured. + * We use the NtQueryInformationFile() system call instead of FindFirstStream() + * and FindNextStream(). This is done for two reasons: * - * Returns 0 on success; nonzero on failure. + * - FindFirstStream() opens its own handle to the file or directory and + * apparently does so without specifying FILE_FLAG_BACKUP_SEMANTICS, thereby + * causing access denied errors on certain files (even when running as the + * Administrator). + * - FindFirstStream() and FindNextStream() is only available on Windows Vista + * and later, whereas the stream support in NtQueryInformationFile() was + * already present in Windows XP. */ static int -win32_capture_streams(const wchar_t *path, - size_t path_num_chars, - struct wim_inode *inode, - struct wim_lookup_table *lookup_table, - u64 file_size, - unsigned vol_flags) +winnt_scan_streams(HANDLE *hFile_p, const wchar_t *path, size_t path_nchars, + struct wim_inode *inode, struct list_head *unhashed_streams, + u64 file_size, u32 vol_flags) { - WIN32_FIND_STREAM_DATA dat; int ret; - HANDLE hFind; - DWORD err; + u8 _buf[1024] _aligned_attribute(8); + u8 *buf; + size_t bufsize; + IO_STATUS_BLOCK iosb; + NTSTATUS status; + const FILE_STREAM_INFORMATION *info; - DEBUG("Capturing streams from \"%ls\"", path); + buf = _buf; + bufsize = sizeof(_buf); - if (win32func_FindFirstStreamW == NULL || - !(vol_flags & FILE_NAMED_STREAMS)) + if (!(vol_flags & FILE_NAMED_STREAMS)) goto unnamed_only; - hFind = win32func_FindFirstStreamW(path, FindStreamInfoStandard, &dat, 0); - if (hFind == INVALID_HANDLE_VALUE) { - err = GetLastError(); - if (err == ERROR_CALL_NOT_IMPLEMENTED) - goto unnamed_only; + /* Get a buffer containing the stream information. */ + while (!NT_SUCCESS(status = (*func_NtQueryInformationFile)(*hFile_p, + &iosb, + buf, + bufsize, + FileStreamInformation))) + { - /* Seems legal for this to return ERROR_HANDLE_EOF on reparse - * points and directories */ - if ((inode->i_attributes & - (FILE_ATTRIBUTE_REPARSE_POINT | FILE_ATTRIBUTE_DIRECTORY)) - && err == ERROR_HANDLE_EOF) - { - DEBUG("ERROR_HANDLE_EOF (ok)"); - return 0; - } else { - if (err == ERROR_ACCESS_DENIED) { - WARNING("Failed to look up data streams " - "of \"%ls\": Access denied!\n%ls", - path, capture_access_denied_msg); - return 0; - } else { - ERROR("Failed to look up data streams " - "of \"%ls\"", path); - win32_error(err); - return WIMLIB_ERR_READ; + switch (status) { + case STATUS_BUFFER_OVERFLOW: + { + u8 *newbuf; + + bufsize *= 2; + if (buf == _buf) + newbuf = MALLOC(bufsize); + else + newbuf = REALLOC(buf, bufsize); + if (!newbuf) { + ret = WIMLIB_ERR_NOMEM; + goto out_free_buf; + } + buf = newbuf; } + break; + case STATUS_NOT_IMPLEMENTED: + case STATUS_NOT_SUPPORTED: + case STATUS_INVALID_INFO_CLASS: + goto unnamed_only; + default: + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Failed to query stream " + "information (status=0x%08"PRIx32")", + printable_path(path), (u32)status); + ret = WIMLIB_ERR_READ; + goto out_free_buf; } } - do { - ret = win32_capture_stream(path, - path_num_chars, - inode, lookup_table, - &dat); + + if (iosb.Information == 0) { + /* No stream information. */ + ret = 0; + goto out_free_buf; + } + + if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) { + /* OpenEncryptedFileRaw() seems to fail with + * ERROR_SHARING_VIOLATION if there are any handles opened to + * the file. */ + (*func_NtClose)(*hFile_p); + *hFile_p = INVALID_HANDLE_VALUE; + } + + /* Parse one or more stream information structures. */ + info = (const FILE_STREAM_INFORMATION *)buf; + for (;;) { + /* Load the stream information. */ + ret = winnt_scan_stream(path, path_nchars, + info->StreamName, + info->StreamNameLength / 2, + info->StreamSize.QuadPart, + inode, unhashed_streams); if (ret) - goto out_find_close; - } while (win32func_FindNextStreamW(hFind, &dat)); - err = GetLastError(); - if (err != ERROR_HANDLE_EOF) { - ERROR("Win32 API: Error reading data streams from \"%ls\"", path); - win32_error(err); - ret = WIMLIB_ERR_READ; + goto out_free_buf; + + if (info->NextEntryOffset == 0) { + /* No more stream information. */ + break; + } + /* Advance to next stream information. */ + info = (const FILE_STREAM_INFORMATION *) + ((const u8 *)info + info->NextEntryOffset); } -out_find_close: - FindClose(hFind); - return ret; + ret = 0; + goto out_free_buf; + unnamed_only: - /* FindFirstStreamW() API is not available, or the volume does not - * support named streams. Only capture the unnamed data stream. */ - DEBUG("Only capturing unnamed data stream"); - if (inode->i_attributes & - (FILE_ATTRIBUTE_REPARSE_POINT | FILE_ATTRIBUTE_DIRECTORY)) + /* The volume does not support named streams. Only capture the unnamed + * data stream. */ + if (inode->i_attributes & (FILE_ATTRIBUTE_DIRECTORY | + FILE_ATTRIBUTE_REPARSE_POINT)) { ret = 0; - } else { - /* Just create our own WIN32_FIND_STREAM_DATA for an unnamed - * stream to reduce the code to a call to the - * already-implemented win32_capture_stream() */ - wcscpy(dat.cStreamName, L"::$DATA"); - dat.StreamSize.QuadPart = file_size; - ret = win32_capture_stream(path, - path_num_chars, - inode, lookup_table, - &dat); + goto out_free_buf; } + + ret = winnt_scan_stream(path, path_nchars, L"::$DATA", 7, + file_size, inode, unhashed_streams); +out_free_buf: + /* Free buffer if allocated on heap. */ + if (unlikely(buf != _buf)) + FREE(buf); return ret; } static int -win32_build_dentry_tree_recursive(struct wim_dentry **root_ret, - wchar_t *path, - size_t path_num_chars, +winnt_build_dentry_tree_recursive(struct wim_dentry **root_ret, + HANDLE cur_dir, + wchar_t *full_path, + size_t full_path_nchars, + const wchar_t *filename, + size_t filename_nchars, struct add_image_params *params, - struct win32_capture_state *state, - unsigned vol_flags) + struct winnt_scan_stats *stats, + u32 vol_flags) { struct wim_dentry *root = NULL; - struct wim_inode *inode; - DWORD err; - u64 file_size; + struct wim_inode *inode = NULL; + HANDLE h = INVALID_HANDLE_VALUE; int ret; + NTSTATUS status; + FILE_ALL_INFORMATION file_info; u8 *rpbuf; u16 rpbuflen; u16 not_rpfixed; - if (exclude_path(path, path_num_chars, params->config, true)) { - if (params->add_flags & WIMLIB_ADD_FLAG_ROOT) { - ERROR("Cannot exclude the root directory from capture"); - ret = WIMLIB_ERR_INVALID_CAPTURE_CONFIG; - goto out; - } - if ((params->add_flags & WIMLIB_ADD_FLAG_EXCLUDE_VERBOSE) - && params->progress_func) - { - union wimlib_progress_info info; - info.scan.cur_path = path; - info.scan.excluded = true; - params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info); - } + if (should_exclude_path(full_path + params->capture_root_nchars, + full_path_nchars - params->capture_root_nchars, + params->config)) + { ret = 0; - goto out; + goto out_progress; } -#if 0 - if (path_num_chars >= 4 && - !wmemcmp(path, L"\\\\?\\", 4) && - path_num_chars + 1 - 4 > MAX_PATH && - state->num_long_path_warnings < MAX_CAPTURE_LONG_PATH_WARNINGS) - { - WARNING("Path \"%ls\" exceeds MAX_PATH", path); - if (++state->num_long_path_warnings == MAX_CAPTURE_LONG_PATH_WARNINGS) - WARNING("Suppressing further warnings about long paths."); + /* Open the file. */ + status = winnt_openat(cur_dir, + (cur_dir ? filename : full_path), + (cur_dir ? filename_nchars : full_path_nchars), + FILE_READ_DATA | + FILE_READ_ATTRIBUTES | + READ_CONTROL | + ACCESS_SYSTEM_SECURITY | + SYNCHRONIZE, + &h); + if (unlikely(!NT_SUCCESS(status))) { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Can't open file " + "(status=0x%08"PRIx32")", + printable_path(full_path), (u32)status); + ret = WIMLIB_ERR_OPEN; + goto out; } -#endif - if ((params->add_flags & WIMLIB_ADD_FLAG_VERBOSE) - && params->progress_func) + /* Get information about the file. */ { - union wimlib_progress_info info; - info.scan.cur_path = path; - info.scan.excluded = false; - params->progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info); - } + IO_STATUS_BLOCK iosb; - HANDLE hFile = win32_open_existing_file(path, - FILE_READ_DATA | FILE_READ_ATTRIBUTES); - if (hFile == INVALID_HANDLE_VALUE) { - err = GetLastError(); - ERROR("Win32 API: Failed to open \"%ls\"", path); - win32_error(err); - ret = WIMLIB_ERR_OPEN; - goto out; + status = (*func_NtQueryInformationFile)(h, &iosb, + &file_info, + sizeof(file_info), + FileAllInformation); + + if (unlikely(!NT_SUCCESS(status) && + status != STATUS_BUFFER_OVERFLOW)) + { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Can't get file information " + "(status=0x%08"PRIx32")", + printable_path(full_path), (u32)status); + ret = WIMLIB_ERR_STAT; + goto out; + } } - BY_HANDLE_FILE_INFORMATION file_info; - if (!GetFileInformationByHandle(hFile, &file_info)) { - err = GetLastError(); - ERROR("Win32 API: Failed to get file information for \"%ls\"", - path); - win32_error(err); - ret = WIMLIB_ERR_STAT; - goto out_close_handle; + if (unlikely(!cur_dir)) { + + /* Root of tree being captured; get volume information. */ + + FILE_FS_ATTRIBUTE_INFORMATION attr_info; + FILE_FS_VOLUME_INFORMATION vol_info; + IO_STATUS_BLOCK iosb; + + /* Get volume flags */ + status = (*func_NtQueryVolumeInformationFile)(h, &iosb, + &attr_info, + sizeof(attr_info), + FileFsAttributeInformation); + if (likely((NT_SUCCESS(status) || + (status == STATUS_BUFFER_OVERFLOW)) && + (iosb.Information >= + offsetof(FILE_FS_ATTRIBUTE_INFORMATION, + FileSystemAttributes) + + sizeof(attr_info.FileSystemAttributes)))) + { + vol_flags = attr_info.FileSystemAttributes; + } else { + set_errno_from_nt_status(status); + WARNING_WITH_ERRNO("\"%ls\": Can't get volume attributes " + "(status=0x%08"PRIx32")", + printable_path(full_path), + (u32)status); + vol_flags = 0; + } + + /* Set inode number of root directory */ + params->capture_root_ino = + file_info.InternalInformation.IndexNumber.QuadPart; + + /* Get volume ID. */ + status = (*func_NtQueryVolumeInformationFile)(h, &iosb, + &vol_info, + sizeof(vol_info), + FileFsVolumeInformation); + if (likely((NT_SUCCESS(status) || + (status == STATUS_BUFFER_OVERFLOW)) && + (iosb.Information >= + offsetof(FILE_FS_VOLUME_INFORMATION, + VolumeSerialNumber) + + sizeof(vol_info.VolumeSerialNumber)))) + { + params->capture_root_dev = vol_info.VolumeSerialNumber; + } else { + set_errno_from_nt_status(status); + WARNING_WITH_ERRNO("\"%ls\": Can't get volume ID " + "(status=0x%08"PRIx32")", + printable_path(full_path), + (u32)status); + params->capture_root_dev = 0; + } } - if (file_info.dwFileAttributes & FILE_ATTRIBUTE_REPARSE_POINT) { + /* If this is a reparse point, read the reparse data. */ + if (unlikely(file_info.BasicInformation.FileAttributes & + FILE_ATTRIBUTE_REPARSE_POINT)) + { rpbuf = alloca(REPARSE_POINT_MAX_SIZE); - ret = win32_get_reparse_data(hFile, path, params, + ret = winnt_get_reparse_data(h, full_path, params, rpbuf, &rpbuflen); - if (ret < 0) { - /* WIMLIB_ERR_* (inverted) */ - ret = -ret; - goto out_close_handle; - } else if (ret & RP_FIXED) { - not_rpfixed = 0; - } else if (ret == RP_EXCLUDED) { + switch (ret) { + case RP_EXCLUDED: ret = 0; - goto out_close_handle; - } else { + goto out; + case RP_FIXED: + not_rpfixed = 0; + break; + case RP_NOT_FIXED: not_rpfixed = 1; + break; + default: + ERROR_WITH_ERRNO("\"%ls\": Can't get reparse data", + printable_path(full_path)); + goto out; } } @@ -1016,81 +1214,132 @@ win32_build_dentry_tree_recursive(struct wim_dentry **root_ret, * only 1 link and refuse to hard link them. This is because Windows * has a bug where it can return duplicate File IDs for files and * directories on the FAT filesystem. */ - ret = inode_table_new_dentry(¶ms->inode_table, - path_basename_with_len(path, path_num_chars), - ((u64)file_info.nFileIndexHigh << 32) | - (u64)file_info.nFileIndexLow, - file_info.dwVolumeSerialNumber, - (file_info.nNumberOfLinks <= 1 || - (file_info.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)), + ret = inode_table_new_dentry(params->inode_table, + filename, + file_info.InternalInformation.IndexNumber.QuadPart, + 0, /* We don't follow mount points, so we + currently don't need to get the + volume ID / device number. */ + (file_info.StandardInformation.NumberOfLinks <= 1 || + (file_info.BasicInformation.FileAttributes & + FILE_ATTRIBUTE_DIRECTORY)), &root); if (ret) - goto out_close_handle; + goto out; - ret = win32_get_short_name(root, path); - if (ret) - goto out_close_handle; + /* Get the short (DOS) name of the file. */ + status = winnt_get_short_name(h, root); + + /* If we can't read the short filename for any reason other than + * out-of-memory, just ignore the error and assume the file has no short + * name. This shouldn't be an issue, since the short names are + * essentially obsolete anyway. */ + if (unlikely(status == STATUS_NO_MEMORY)) { + ret = WIMLIB_ERR_NOMEM; + goto out; + } inode = root->d_inode; - if (inode->i_nlink > 1) /* Shared inode; nothing more to do */ - goto out_close_handle; + if (inode->i_nlink > 1) { + /* Shared inode (hard link); skip reading per-inode information. + */ + ret = 0; + goto out_progress; + } - inode->i_attributes = file_info.dwFileAttributes; - inode->i_creation_time = FILETIME_to_u64(&file_info.ftCreationTime); - inode->i_last_write_time = FILETIME_to_u64(&file_info.ftLastWriteTime); - inode->i_last_access_time = FILETIME_to_u64(&file_info.ftLastAccessTime); + inode->i_attributes = file_info.BasicInformation.FileAttributes; + inode->i_creation_time = file_info.BasicInformation.CreationTime.QuadPart; + inode->i_last_write_time = file_info.BasicInformation.LastWriteTime.QuadPart; + inode->i_last_access_time = file_info.BasicInformation.LastAccessTime.QuadPart; inode->i_resolved = 1; - params->add_flags &= ~WIMLIB_ADD_FLAG_ROOT; - + /* Get the file's security descriptor, unless we are capturing in + * NO_ACLS mode or the volume does not support security descriptors. */ if (!(params->add_flags & WIMLIB_ADD_FLAG_NO_ACLS) && (vol_flags & FILE_PERSISTENT_ACLS)) { - ret = win32_get_security_descriptor(root, ¶ms->sd_set, - path, state, - params->add_flags); - if (ret) - goto out_close_handle; + status = winnt_get_security_descriptor(h, inode, + params->sd_set, stats, + params->add_flags); + if (!NT_SUCCESS(status)) { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Can't read security " + "descriptor (status=0x%08"PRIu32")", + printable_path(full_path), + (u32)status); + ret = WIMLIB_ERR_STAT; + goto out; + } } - file_size = ((u64)file_info.nFileSizeHigh << 32) | - (u64)file_info.nFileSizeLow; - - CloseHandle(hFile); - - /* Capture the unnamed data stream (only should be present for regular - * files) and any alternate data streams. */ - ret = win32_capture_streams(path, - path_num_chars, - inode, - params->lookup_table, - file_size, - vol_flags); + /* Load information about the unnamed data stream and any named data + * streams. */ + ret = winnt_scan_streams(&h, + full_path, + full_path_nchars, + inode, + params->unhashed_streams, + file_info.StandardInformation.EndOfFile.QuadPart, + vol_flags); if (ret) goto out; - if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) { - /* Reparse point: set the reparse data (which we read already) - * */ + if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)) { + + /* Reparse point: set the reparse data (already read). */ + inode->i_not_rpfixed = not_rpfixed; inode->i_reparse_tag = le32_to_cpu(*(le32*)rpbuf); ret = inode_set_unnamed_stream(inode, rpbuf + 8, rpbuflen - 8, params->lookup_table); + if (ret) + goto out; } else if (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY) { - /* Directory (not a reparse point) --- recurse to children */ - ret = win32_recurse_directory(root, - path, - path_num_chars, + + /* Directory: recurse to children. */ + + if (unlikely(h == INVALID_HANDLE_VALUE)) { + /* Re-open handle that was closed to read raw encrypted + * data. */ + status = winnt_openat(cur_dir, + (cur_dir ? + filename : full_path), + (cur_dir ? + filename_nchars : full_path_nchars), + FILE_LIST_DIRECTORY | SYNCHRONIZE, + &h); + if (!NT_SUCCESS(status)) { + set_errno_from_nt_status(status); + ERROR_WITH_ERRNO("\"%ls\": Can't re-open file " + "(status=0x%08"PRIx32")", + printable_path(full_path), + (u32)status); + ret = WIMLIB_ERR_OPEN; + goto out; + } + } + ret = winnt_recurse_directory(h, + full_path, + full_path_nchars, + root, params, - state, + stats, vol_flags); + if (ret) + goto out; } - goto out; -out_close_handle: - CloseHandle(hFile); + +out_progress: + params->progress.scan.cur_path = printable_path(full_path); + if (likely(root)) + ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_OK, inode); + else + ret = do_capture_progress(params, WIMLIB_SCAN_DENTRY_EXCLUDED, NULL); out: - if (ret == 0) + if (likely(h != INVALID_HANDLE_VALUE)) + (*func_NtClose)(h); + if (likely(ret == 0)) *root_ret = root; else free_dentry_tree(root, params->lookup_table); @@ -1098,125 +1347,78 @@ out: } static void -win32_do_capture_warnings(const struct win32_capture_state *state, - int add_flags) +winnt_do_scan_warnings(const wchar_t *path, const struct winnt_scan_stats *stats) { - if (state->num_get_sacl_priv_notheld == 0 && - state->num_get_sd_access_denied == 0) + if (likely(stats->num_get_sacl_priv_notheld == 0 && + stats->num_get_sd_access_denied == 0)) return; - WARNING(""); - WARNING("Built dentry tree successfully, but with the following problem(s):"); - if (state->num_get_sacl_priv_notheld != 0) { - WARNING("Could not capture SACL (System Access Control List)\n" - " on %lu files or directories.", - state->num_get_sacl_priv_notheld); + WARNING("Scan of \"%ls\" complete, but with one or more warnings:", path); + if (stats->num_get_sacl_priv_notheld != 0) { + WARNING("- Could not capture SACL (System Access Control List)\n" + " on %lu files or directories.", + stats->num_get_sacl_priv_notheld); } - if (state->num_get_sd_access_denied != 0) { - WARNING("Could not capture security descriptor at all\n" - " on %lu files or directories.", - state->num_get_sd_access_denied); + if (stats->num_get_sd_access_denied != 0) { + WARNING("- Could not capture security descriptor at all\n" + " on %lu files or directories.", + stats->num_get_sd_access_denied); } - WARNING( - "Try running the program as the Administrator to make sure all the\n" -" desired metadata has been captured exactly. However, if you\n" -" do not care about capturing security descriptors correctly, then\n" -" nothing more needs to be done%ls\n", - (add_flags & WIMLIB_ADD_FLAG_NO_ACLS) ? L"." : - L", although you might consider\n" -" using the --no-acls option to explicitly capture no security\n" -" descriptors.\n"); + WARNING("To fully capture all security descriptors, run the program\n" + " with Administrator rights."); } #define WINDOWS_NT_MAX_PATH 32768 -/* Win32 version of capturing a directory tree */ +/* Win32 version of capturing a directory tree. */ int win32_build_dentry_tree(struct wim_dentry **root_ret, const wchar_t *root_disk_path, struct add_image_params *params) { - size_t path_nchars; wchar_t *path; - int ret; - struct win32_capture_state state; - unsigned vol_flags; DWORD dret; - bool need_prefix_free = false; - - if (!win32func_FindFirstStreamW) { - WARNING("Running on Windows XP or earlier; " - "alternate data streams will not be captured."); - } - - path_nchars = wcslen(root_disk_path); - if (path_nchars > WINDOWS_NT_MAX_PATH) - return WIMLIB_ERR_INVALID_PARAM; - - if (GetFileAttributesW(root_disk_path) == INVALID_FILE_ATTRIBUTES && - GetLastError() == ERROR_FILE_NOT_FOUND) - { - ERROR("Capture directory \"%ls\" does not exist!", - root_disk_path); - return WIMLIB_ERR_OPENDIR; - } - - ret = win32_get_file_and_vol_ids(root_disk_path, - ¶ms->capture_root_ino, - ¶ms->capture_root_dev); - if (ret) - return ret; - - win32_get_vol_flags(root_disk_path, &vol_flags); + size_t path_nchars; + int ret; + struct winnt_scan_stats stats; /* WARNING: There is no check for overflow later when this buffer is * being used! But it's as long as the maximum path length understood - * by Windows NT (which is NOT the same as MAX_PATH). */ - path = MALLOC(WINDOWS_NT_MAX_PATH * sizeof(wchar_t)); + * by Windows NT (which is NOT the same as MAX_PATH). */ + path = MALLOC((WINDOWS_NT_MAX_PATH + 1) * sizeof(wchar_t)); if (!path) return WIMLIB_ERR_NOMEM; - /* Work around defective behavior in Windows where paths longer than 260 - * characters are not supported by default; instead they need to be - * turned into absolute paths and prefixed with "\\?\". */ + /* Translate into full path. */ + dret = GetFullPathName(root_disk_path, WINDOWS_NT_MAX_PATH - 3, + &path[4], NULL); - if (wcsncmp(root_disk_path, L"\\\\?\\", 4)) { - dret = GetFullPathName(root_disk_path, WINDOWS_NT_MAX_PATH - 4, - &path[4], NULL); - - if (dret == 0 || dret >= WINDOWS_NT_MAX_PATH - 4) { - WARNING("Can't get full path name for \"%ls\"", root_disk_path); - wmemcpy(path, root_disk_path, path_nchars + 1); - } else { - wmemcpy(path, L"\\\\?\\", 4); - path_nchars = 4 + dret; - /* Update pattern prefix */ - if (params->config != NULL) - { - params->config->_prefix = TSTRDUP(path); - params->config->_prefix_num_tchars = path_nchars; - if (params->config->_prefix == NULL) - { - ret = WIMLIB_ERR_NOMEM; - goto out_free_path; - } - need_prefix_free = true; - } - } - } else { - wmemcpy(path, root_disk_path, path_nchars + 1); + if (unlikely(dret == 0 || dret >= WINDOWS_NT_MAX_PATH - 3)) { + ERROR("Can't get full path name for \"%ls\"", root_disk_path); + return WIMLIB_ERR_UNSUPPORTED; } - memset(&state, 0, sizeof(state)); - ret = win32_build_dentry_tree_recursive(root_ret, path, - path_nchars, params, - &state, vol_flags); - if (need_prefix_free) - FREE(params->config->_prefix); -out_free_path: + /* Add \??\ prefix to form the NT namespace path. */ + wmemcpy(path, L"\\??\\", 4); + path_nchars = dret + 4; + + /* Strip trailing slashes. If we don't do this, we may create a path + * with multiple consecutive backslashes, which for some reason causes + * Windows to report that the file cannot be found. */ + while (unlikely(path[path_nchars - 1] == L'\\' && + path[path_nchars - 2] != L':')) + path[--path_nchars] = L'\0'; + + params->capture_root_nchars = path_nchars; + + memset(&stats, 0, sizeof(stats)); + + ret = winnt_build_dentry_tree_recursive(root_ret, NULL, + path, path_nchars, L"", 0, + params, &stats, 0); FREE(path); if (ret == 0) - win32_do_capture_warnings(&state, params->add_flags); + winnt_do_scan_warnings(root_disk_path, &stats); return ret; }