X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fdentry.c;h=b0ebf64dab19a10d81350c10f3f0386eb849aa2c;hp=a9d13f690d01afffde23d30f53017279b5ff19ff;hb=bec8ad2db6634b814024f014ace69b0301fd5032;hpb=9b1647d673092a6c53a90c1d89831a5131660b9d diff --git a/src/dentry.c b/src/dentry.c index a9d13f69..378e16fe 100644 --- a/src/dentry.c +++ b/src/dentry.c @@ -1,1659 +1,1936 @@ /* - * dentry.c - * - * A dentry (directory entry) contains the metadata for a file. In the WIM file - * format, the dentries are stored in the "metadata resource" section right - * after the security data. Each image in the WIM file has its own metadata - * resource with its own security data and dentry tree. Dentries in different - * images may share file resources by referring to the same lookup table - * entries. + * dentry.c - see description below */ /* - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers * - * This file is part of wimlib, a library for working with WIM files. + * This file is free software; you can redistribute it and/or modify it under + * the terms of the GNU Lesser General Public License as published by the Free + * Software Foundation; either version 3 of the License, or (at your option) any + * later version. * - * wimlib is free software; you can redistribute it and/or modify it under the - * terms of the GNU General Public License as published by the Free Software - * Foundation; either version 3 of the License, or (at your option) any later - * version. + * This file is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS + * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more + * details. * - * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY - * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR - * A PARTICULAR PURPOSE. See the GNU General Public License for more details. + * You should have received a copy of the GNU Lesser General Public License + * along with this file; if not, see http://www.gnu.org/licenses/. + */ + +/* + * This file contains logic to deal with WIM directory entries, or "dentries": + * + * - Reading a dentry tree from a metadata resource in a WIM file + * - Writing a dentry tree to a metadata resource in a WIM file + * - Iterating through a tree of WIM dentries + * - Path lookup: translating a path into a WIM dentry or inode + * - Creating, modifying, and deleting WIM dentries + * + * Notes: + * + * - A WIM file can contain multiple images, each of which has an independent + * tree of dentries. "On disk", the dentry tree for an image is stored in + * the "metadata resource" for that image. + * + * - Multiple dentries in an image may correspond to the same inode, or "file". + * When this occurs, it means that the file has multiple names, or "hard + * links". A dentry is not a file, but rather the name of a file! + * + * - Inodes are not represented explicitly in the WIM file format. Instead, + * the metadata resource provides a "hard link group ID" for each dentry. + * wimlib handles pulling out actual inodes from this information, but this + * occurs in inode_fixup.c and not in this file. * - * You should have received a copy of the GNU General Public License along with - * wimlib; if not, see http://www.gnu.org/licenses/. + * - wimlib does not allow *directory* hard links, so a WIM image really does + * have a *tree* of dentries (and not an arbitrary graph of dentries). + * + * - wimlib indexes dentries both case-insensitively and case-sensitively, + * allowing either behavior to be used for path lookup. + * + * - Multiple dentries in a directory might have the same case-insensitive + * name. But wimlib enforces that at most one dentry in a directory can have + * a given case-sensitive name. */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + #include -#include -#include -#include -#include "dentry.h" -#include "io.h" -#include "lookup_table.h" -#include "sha1.h" -#include "timestamp.h" -#include "wimlib_internal.h" +#include "wimlib/assert.h" +#include "wimlib/dentry.h" +#include "wimlib/inode.h" +#include "wimlib/encoding.h" +#include "wimlib/endianness.h" +#include "wimlib/metadata.h" +#include "wimlib/paths.h" + +/* On-disk format of a WIM dentry (directory entry), located in the metadata + * resource for a WIM image. */ +struct wim_dentry_on_disk { + + /* Length of this directory entry in bytes, not including any extra + * stream entries. Should be a multiple of 8 so that the following + * dentry or extra stream entry is aligned on an 8-byte boundary. (If + * not, wimlib will round it up.) It must be at least as long as the + * fixed-length fields of the dentry (WIM_DENTRY_DISK_SIZE), plus the + * lengths of the file name and/or short name if present, plus the size + * of any "extra" data. + * + * It is also possible for this field to be 0. This case indicates the + * end of a list of sibling entries in a directory. It also means the + * real length is 8, because the dentry included only the length field, + * but that takes up 8 bytes. */ + le64 length; + + /* File attributes for the file or directory. This is a bitwise OR of + * the FILE_ATTRIBUTE_* constants and should correspond to the value + * retrieved by GetFileAttributes() on Windows. */ + le32 attributes; + + /* A value that specifies the security descriptor for this file or + * directory. If -1, the file or directory has no security descriptor. + * Otherwise, it is a 0-based index into the WIM image's table of + * security descriptors (see: `struct wim_security_data') */ + sle32 security_id; + + /* Offset, in bytes, from the start of the uncompressed metadata + * resource of this directory's child directory entries, or 0 if this + * directory entry does not correspond to a directory or otherwise does + * not have any children. */ + le64 subdir_offset; + + /* Reserved fields */ + le64 unused_1; + le64 unused_2; + + /* Creation time, last access time, and last write time, in + * 100-nanosecond intervals since 12:00 a.m UTC January 1, 1601. They + * should correspond to the times gotten by calling GetFileTime() on + * Windows. */ + le64 creation_time; + le64 last_access_time; + le64 last_write_time; -/* - * Returns true if @dentry has the UTF-8 file name @name that has length - * @name_len. - */ -static bool dentry_has_name(const struct dentry *dentry, const char *name, - size_t name_len) -{ - if (dentry->file_name_utf8_len != name_len) - return false; - return memcmp(dentry->file_name_utf8, name, name_len) == 0; -} + /* + * Usually this is the SHA-1 message digest of the file's "contents" + * (the unnamed data stream). + * + * If the file has FILE_ATTRIBUTE_REPARSE_POINT set, then this is + * instead usually the SHA-1 message digest of the uncompressed reparse + * point data. + * + * However, there are some special rules that need to be applied to + * interpret this field correctly when extra stream entries are present. + * See the code for details. + */ + u8 default_hash[SHA1_HASH_SIZE]; -static u64 __dentry_correct_length_unaligned(u16 file_name_len, - u16 short_name_len) -{ - u64 length = WIM_DENTRY_DISK_SIZE; - if (file_name_len) - length += file_name_len + 2; - if (short_name_len) - length += short_name_len + 2; - return length; -} + /* The format of the following data is not yet completely known and they + * do not correspond to Microsoft's documentation. + * + * If this directory entry is for a reparse point (has + * FILE_ATTRIBUTE_REPARSE_POINT set in the 'attributes' field), then the + * version of the following fields containing the reparse tag is valid. + * Furthermore, the field notated as not_rpfixed, as far as I can tell, + * is supposed to be set to 1 if reparse point fixups (a.k.a. fixing the + * targets of absolute symbolic links) were *not* done, and otherwise 0. + * + * If this directory entry is not for a reparse point, then the version + * of the following fields containing the hard_link_group_id is valid. + * All MS says about this field is that "If this file is part of a hard + * link set, all the directory entries in the set will share the same + * value in this field.". However, more specifically I have observed + * the following: + * - If the file is part of a hard link set of size 1, then the + * hard_link_group_id should be set to either 0, which is treated + * specially as indicating "not hardlinked", or any unique value. + * - The specific nonzero values used to identity hard link sets do + * not matter, as long as they are unique. + * - However, due to bugs in Microsoft's software, it is actually NOT + * guaranteed that directory entries that share the same hard link + * group ID are actually hard linked to each either. See + * inode_fixup.c for the code that handles this. + */ + union { + struct { + le32 rp_unknown_1; + le32 reparse_tag; + le16 rp_unknown_2; + le16 not_rpfixed; + } _packed_attribute reparse; + struct { + le32 rp_unknown_1; + le64 hard_link_group_id; + } _packed_attribute nonreparse; + }; + + /* Number of extra stream entries that directly follow this dentry + * on-disk. */ + le16 num_extra_streams; + + /* If nonzero, this is the length, in bytes, of this dentry's UTF-16LE + * encoded short name (8.3 DOS-compatible name), excluding the null + * terminator. If zero, then the long name of this dentry does not have + * a corresponding short name (but this does not exclude the possibility + * that another dentry for the same file has a short name). */ + le16 short_name_nbytes; + + /* If nonzero, this is the length, in bytes, of this dentry's UTF-16LE + * encoded "long" name, excluding the null terminator. If zero, then + * this file has no long name. The root dentry should not have a long + * name, but all other dentries in the image should have long names. */ + le16 file_name_nbytes; + + /* Beginning of optional, variable-length fields */ + + /* If file_name_nbytes != 0, the next field will be the UTF-16LE encoded + * long file name. This will be null-terminated, so the size of this + * field will really be file_name_nbytes + 2. */ + /*utf16lechar file_name[];*/ + + /* If short_name_nbytes != 0, the next field will be the UTF-16LE + * encoded short name. This will be null-terminated, so the size of + * this field will really be short_name_nbytes + 2. */ + /*utf16lechar short_name[];*/ + + /* If there is still space in the dentry (according to the 'length' + * field) after 8-byte alignment, then the remaining space will be a + * variable-length list of tagged metadata items. See tagged_items.c + * for more information. */ + /* u8 tagged_items[] _aligned_attribute(8); */ + +} _packed_attribute; + /* If num_extra_streams != 0, then there are that many extra stream + * entries following the dentry, starting on the next 8-byte aligned + * boundary. They are not counted in the 'length' field of the dentry. + */ -static u64 dentry_correct_length_unaligned(const struct dentry *dentry) -{ - return __dentry_correct_length_unaligned(dentry->file_name_len, - dentry->short_name_len); -} +/* On-disk format of an extra stream entry. This represents an extra NTFS-style + * "stream" associated with the file, such as a named data stream. */ +struct wim_extra_stream_entry_on_disk { -/* Return the "correct" value to write in the length field of the dentry, based - * on the file name length and short name length */ -static u64 dentry_correct_length(const struct dentry *dentry) -{ - return (dentry_correct_length_unaligned(dentry) + 7) & ~7; -} + /* Length of this extra stream entry, in bytes. This includes all + * fixed-length fields, plus the name and null terminator if present, + * and any needed padding such that the length is a multiple of 8. */ + le64 length; -static u64 __dentry_total_length(const struct dentry *dentry, u64 length) -{ - const struct inode *inode = dentry->inode; - for (u16 i = 0; i < inode->num_ads; i++) - length += ads_entry_total_length(inode->ads_entries[i]); - return (length + 7) & ~7; -} + /* Reserved field */ + le64 reserved; -u64 dentry_correct_total_length(const struct dentry *dentry) -{ - return __dentry_total_length(dentry, - dentry_correct_length_unaligned(dentry)); -} + /* SHA-1 message digest of this stream's uncompressed data, or all + * zeroes if this stream's data is of zero length. */ + u8 hash[SHA1_HASH_SIZE]; -/* Real length of a dentry, including the alternate data stream entries, which - * are not included in the dentry->length field... */ -u64 dentry_total_length(const struct dentry *dentry) + /* Length of this stream's name, in bytes and excluding the null + * terminator; or 0 if this stream is unnamed. */ + le16 name_nbytes; + + /* Stream name in UTF-16LE. It is @name_nbytes bytes long, excluding + * the null terminator. There is a null terminator character if + * @name_nbytes != 0; i.e., if this stream is named. */ + utf16lechar name[]; +} _packed_attribute; + +static void +do_dentry_set_name(struct wim_dentry *dentry, utf16lechar *file_name, + size_t file_name_nbytes) { - return __dentry_total_length(dentry, dentry->length); + FREE(dentry->file_name); + dentry->file_name = file_name; + dentry->file_name_nbytes = file_name_nbytes; + + if (dentry_has_short_name(dentry)) { + FREE(dentry->short_name); + dentry->short_name = NULL; + dentry->short_name_nbytes = 0; + } } -/* Transfers file attributes from a `stat' buffer to an inode. */ -void stbuf_to_inode(const struct stat *stbuf, struct inode *inode) +/* + * Set the name of a WIM dentry from a UTF-16LE string. + * + * This sets the long name of the dentry. The short name will automatically be + * removed, since it may not be appropriate for the new long name. + * + * The @name string need not be null-terminated, since its length is specified + * in @name_nbytes. + * + * If @name_nbytes is 0, both the long and short names of the dentry will be + * removed. + * + * Only use this function on unlinked dentries, since it doesn't update the name + * indices. For dentries that are currently linked into the tree, use + * rename_wim_path(). + * + * Returns 0 or WIMLIB_ERR_NOMEM. + */ +int +dentry_set_name_utf16le(struct wim_dentry *dentry, const utf16lechar *name, + size_t name_nbytes) { - if (S_ISLNK(stbuf->st_mode)) { - inode->attributes = FILE_ATTRIBUTE_REPARSE_POINT; - inode->reparse_tag = WIM_IO_REPARSE_TAG_SYMLINK; - } else if (S_ISDIR(stbuf->st_mode)) { - inode->attributes = FILE_ATTRIBUTE_DIRECTORY; - } else { - inode->attributes = FILE_ATTRIBUTE_NORMAL; + utf16lechar *dup = NULL; + + if (name_nbytes) { + dup = utf16le_dupz(name, name_nbytes); + if (!dup) + return WIMLIB_ERR_NOMEM; } - if (sizeof(ino_t) >= 8) - inode->ino = (u64)stbuf->st_ino; - else - inode->ino = (u64)stbuf->st_ino | - ((u64)stbuf->st_dev << (sizeof(ino_t) * 8)); - /* Set timestamps */ - inode->creation_time = timespec_to_wim_timestamp(&stbuf->st_mtim); - inode->last_write_time = timespec_to_wim_timestamp(&stbuf->st_mtim); - inode->last_access_time = timespec_to_wim_timestamp(&stbuf->st_atim); + do_dentry_set_name(dentry, dup, name_nbytes); + return 0; } -/* Sets all the timestamp fields of the dentry to the current time. */ -void inode_update_all_timestamps(struct inode *inode) +/* + * Set the name of a WIM dentry from a 'tchar' string. + * + * This sets the long name of the dentry. The short name will automatically be + * removed, since it may not be appropriate for the new long name. + * + * If @name is NULL or empty, both the long and short names of the dentry will + * be removed. + * + * Only use this function on unlinked dentries, since it doesn't update the name + * indices. For dentries that are currently linked into the tree, use + * rename_wim_path(). + * + * Returns 0 or an error code resulting from a failed string conversion. + */ +int +dentry_set_name(struct wim_dentry *dentry, const tchar *name) { - u64 now = get_wim_timestamp(); - inode->creation_time = now; - inode->last_access_time = now; - inode->last_write_time = now; + utf16lechar *name_utf16le = NULL; + size_t name_utf16le_nbytes = 0; + int ret; + + if (name && *name) { + ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar), + &name_utf16le, &name_utf16le_nbytes); + if (ret) + return ret; + } + + do_dentry_set_name(dentry, name_utf16le, name_utf16le_nbytes); + return 0; } -/* Returns the alternate data stream entry belonging to @inode that has the - * stream name @stream_name. */ -struct ads_entry *inode_get_ads_entry(struct inode *inode, - const char *stream_name, - u16 *idx_ret) +/* Calculate the minimum unaligned length, in bytes, of an on-disk WIM dentry + * that has names of the specified lengths. (Zero length means the + * corresponding name actually does not exist.) The returned value excludes + * tagged metadata items as well as any extra stream entries that may need to + * follow the dentry. */ +static size_t +dentry_min_len_with_names(u16 file_name_nbytes, u16 short_name_nbytes) { - size_t stream_name_len; - if (!stream_name) - return NULL; - if (inode->num_ads) { - u16 i = 0; - stream_name_len = strlen(stream_name); - do { - if (ads_entry_has_name(inode->ads_entries[i], - stream_name, stream_name_len)) - { - if (idx_ret) - *idx_ret = i; - return inode->ads_entries[i]; - } - } while (++i != inode->num_ads); - } - return NULL; + size_t length = sizeof(struct wim_dentry_on_disk); + if (file_name_nbytes) + length += (u32)file_name_nbytes + 2; + if (short_name_nbytes) + length += (u32)short_name_nbytes + 2; + return length; } -static struct ads_entry *new_ads_entry(const char *name) +/* Return the length, in bytes, required for the specified stream on-disk, when + * represented as an extra stream entry. */ +static size_t +stream_out_total_length(const struct wim_inode_stream *strm) { - struct ads_entry *ads_entry = CALLOC(1, sizeof(struct ads_entry)); - if (!ads_entry) - return NULL; - if (name && *name) { - if (change_ads_name(ads_entry, name)) { - FREE(ads_entry); - return NULL; - } - } - return ads_entry; + /* Account for the fixed length portion */ + size_t len = sizeof(struct wim_extra_stream_entry_on_disk); + + /* For named streams, account for the variable-length name. */ + if (stream_is_named(strm)) + len += utf16le_len_bytes(strm->stream_name) + 2; + + /* Account for any necessary padding to the next 8-byte boundary. */ + return (len + 7) & ~7; } -/* - * Add an alternate stream entry to an inode and return a pointer to it, or NULL - * if memory could not be allocated. +/* + * Calculate the total number of bytes that will be consumed when a dentry is + * written. This includes the fixed-length portion of the dentry, the name + * fields, any tagged metadata items, and any extra stream entries. This also + * includes all alignment bytes. */ -struct ads_entry *inode_add_ads(struct inode *inode, const char *stream_name) +size_t +dentry_out_total_length(const struct wim_dentry *dentry) { - u16 num_ads; - struct ads_entry **ads_entries; - struct ads_entry *new_entry; + const struct wim_inode *inode = dentry->d_inode; + size_t len; - if (inode->num_ads >= 0xfffe) { - ERROR("Too many alternate data streams in one inode!"); - return NULL; + len = dentry_min_len_with_names(dentry->file_name_nbytes, + dentry->short_name_nbytes); + len = (len + 7) & ~7; + + if (inode->i_extra_size) { + len += inode->i_extra_size; + len = (len + 7) & ~7; } - num_ads = inode->num_ads + 1; - ads_entries = REALLOC(inode->ads_entries, - num_ads * sizeof(inode->ads_entries[0])); - if (!ads_entries) { - ERROR("Failed to allocate memory for new alternate data stream"); - return NULL; + + if (!(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) { + /* + * Extra stream entries: + * + * - Use one extra stream entry for each named data stream + * - Use one extra stream entry for the unnamed data stream when there is either: + * - a reparse point stream + * - at least one named data stream (for Windows PE bug workaround) + * - Use one extra stream entry for the reparse point stream if there is one + */ + bool have_named_data_stream = false; + bool have_reparse_point_stream = false; + for (unsigned i = 0; i < inode->i_num_streams; i++) { + const struct wim_inode_stream *strm = &inode->i_streams[i]; + if (stream_is_named_data_stream(strm)) { + len += stream_out_total_length(strm); + have_named_data_stream = true; + } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) { + wimlib_assert(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT); + have_reparse_point_stream = true; + } + } + + if (have_named_data_stream || have_reparse_point_stream) { + if (have_reparse_point_stream) + len += (sizeof(struct wim_extra_stream_entry_on_disk) + 7) & ~7; + len += (sizeof(struct wim_extra_stream_entry_on_disk) + 7) & ~7; + } } - inode->ads_entries = ads_entries; - new_entry = new_ads_entry(stream_name); - if (new_entry) - return NULL; - inode->num_ads = num_ads; - ads_entries[num_ads - 1] = new_entry; - return new_entry; + return len; } - -/* - * Calls a function on all directory entries in a directory tree. It is called - * on a parent before its children. - */ -int for_dentry_in_tree(struct dentry *root, - int (*visitor)(struct dentry*, void*), void *arg) +/* Internal version of for_dentry_in_tree() that omits the NULL check */ +static int +do_for_dentry_in_tree(struct wim_dentry *dentry, + int (*visitor)(struct wim_dentry *, void *), void *arg) { int ret; - struct dentry *child; + struct wim_dentry *child; - ret = visitor(root, arg); - - if (ret != 0) + ret = (*visitor)(dentry, arg); + if (unlikely(ret)) return ret; - child = root->inode->children; - - if (!child) - return 0; - - do { - ret = for_dentry_in_tree(child, visitor, arg); - if (ret != 0) + for_dentry_child(child, dentry) { + ret = do_for_dentry_in_tree(child, visitor, arg); + if (unlikely(ret)) return ret; - child = child->next; - } while (child != root->inode->children); + } return 0; } -/* - * Like for_dentry_in_tree(), but the visitor function is always called on a - * dentry's children before on itself. - */ -int for_dentry_in_tree_depth(struct dentry *root, - int (*visitor)(struct dentry*, void*), void *arg) +/* Internal version of for_dentry_in_tree_depth() that omits the NULL check */ +static int +do_for_dentry_in_tree_depth(struct wim_dentry *dentry, + int (*visitor)(struct wim_dentry *, void *), void *arg) { int ret; - struct dentry *child; - struct dentry *next; + struct wim_dentry *child; - child = root->inode->children; - if (child) { - do { - next = child->next; - ret = for_dentry_in_tree_depth(child, visitor, arg); - if (ret != 0) - return ret; - child = next; - } while (child != root->inode->children); + for_dentry_child_postorder(child, dentry) { + ret = do_for_dentry_in_tree_depth(child, visitor, arg); + if (unlikely(ret)) + return ret; } - return visitor(root, arg); + return unlikely((*visitor)(dentry, arg)); } -/* - * Calculate the full path of @dentry, based on its parent's full path and on - * its UTF-8 file name. +/* + * Call a function on all dentries in a tree. + * + * @arg will be passed as the second argument to each invocation of @visitor. + * + * This function does a pre-order traversal --- that is, a parent will be + * visited before its children. It also will visit siblings in order of + * case-sensitive filename. Equivalently, this function visits the entire tree + * in the case-sensitive lexicographic order of the full paths. + * + * It is safe to pass NULL for @root, which means that the dentry tree is empty. + * In this case, this function does nothing. + * + * @visitor must not modify the structure of the dentry tree during the + * traversal. + * + * The return value will be 0 if all calls to @visitor returned 0. Otherwise, + * the return value will be the first nonzero value returned by @visitor. */ -int calculate_dentry_full_path(struct dentry *dentry, void *ignore) +int +for_dentry_in_tree(struct wim_dentry *root, + int (*visitor)(struct wim_dentry *, void *), void *arg) { - char *full_path; - u32 full_path_len; - if (dentry_is_root(dentry)) { - full_path = MALLOC(2); - if (!full_path) - goto oom; - full_path[0] = '/'; - full_path[1] = '\0'; - full_path_len = 1; - } else { - char *parent_full_path; - u32 parent_full_path_len; - const struct dentry *parent = dentry->parent; + if (unlikely(!root)) + return 0; + return do_for_dentry_in_tree(root, visitor, arg); +} - if (dentry_is_root(parent)) { - parent_full_path = ""; - parent_full_path_len = 0; - } else { - parent_full_path = parent->full_path_utf8; - parent_full_path_len = parent->full_path_utf8_len; - } +/* Like for_dentry_in_tree(), but do a depth-first traversal of the dentry tree. + * That is, the visitor function will be called on a dentry's children before + * itself. It will be safe to free a dentry when visiting it. */ +static int +for_dentry_in_tree_depth(struct wim_dentry *root, + int (*visitor)(struct wim_dentry *, void *), void *arg) +{ + if (unlikely(!root)) + return 0; + return do_for_dentry_in_tree_depth(root, visitor, arg); +} - full_path_len = parent_full_path_len + 1 + - dentry->file_name_utf8_len; - full_path = MALLOC(full_path_len + 1); - if (!full_path) - goto oom; - - memcpy(full_path, parent_full_path, parent_full_path_len); - full_path[parent_full_path_len] = '/'; - memcpy(full_path + parent_full_path_len + 1, - dentry->file_name_utf8, - dentry->file_name_utf8_len); - full_path[full_path_len] = '\0'; - } - FREE(dentry->full_path_utf8); - dentry->full_path_utf8 = full_path; - dentry->full_path_utf8_len = full_path_len; - return 0; -oom: - ERROR("Out of memory while calculating dentry full path"); - return WIMLIB_ERR_NOMEM; +/* + * Calculate the full path to @dentry within the WIM image, if not already done. + * + * The full name will be saved in the cached value 'dentry->_full_path'. + * + * Whenever possible, use dentry_full_path() instead of calling this and + * accessing _full_path directly. + * + * Returns 0 or an error code resulting from a failed string conversion. + */ +int +calculate_dentry_full_path(struct wim_dentry *dentry) +{ + size_t ulen; + size_t dummy; + const struct wim_dentry *d; + + if (dentry->_full_path) + return 0; + + ulen = 0; + d = dentry; + do { + ulen += d->file_name_nbytes / sizeof(utf16lechar); + ulen++; + d = d->d_parent; /* assumes d == d->d_parent for root */ + } while (!dentry_is_root(d)); + + utf16lechar ubuf[ulen]; + utf16lechar *p = &ubuf[ulen]; + + d = dentry; + do { + p -= d->file_name_nbytes / sizeof(utf16lechar); + memcpy(p, d->file_name, d->file_name_nbytes); + *--p = cpu_to_le16(WIM_PATH_SEPARATOR); + d = d->d_parent; /* assumes d == d->d_parent for root */ + } while (!dentry_is_root(d)); + + wimlib_assert(p == ubuf); + + return utf16le_to_tstr(ubuf, ulen * sizeof(utf16lechar), + &dentry->_full_path, &dummy); } -/* - * Recursively calculates the subdir offsets for a directory tree. +/* + * Return the full path to the @dentry within the WIM image, or NULL if the full + * path could not be determined due to a string conversion error. * - * @dentry: The root of the directory tree. - * @subdir_offset_p: The current subdirectory offset; i.e., the subdirectory - * offset for @dentry. + * The returned memory will be cached in the dentry, so the caller is not + * responsible for freeing it. */ -void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p) +tchar * +dentry_full_path(struct wim_dentry *dentry) +{ + calculate_dentry_full_path(dentry); + return dentry->_full_path; +} + +static int +dentry_calculate_subdir_offset(struct wim_dentry *dentry, void *_subdir_offset_p) { - struct dentry *child; + if (dentry_is_directory(dentry)) { + u64 *subdir_offset_p = _subdir_offset_p; + struct wim_dentry *child; - child = dentry->inode->children; - dentry->subdir_offset = *subdir_offset_p; + /* Set offset of directory's child dentries */ + dentry->subdir_offset = *subdir_offset_p; - if (child) { - /* Advance the subdir offset by the amount of space the children - * of this dentry take up. */ - do { - *subdir_offset_p += dentry_correct_total_length(child); - child = child->next; - } while (child != dentry->inode->children); + /* Account for child dentries */ + for_dentry_child(child, dentry) + *subdir_offset_p += dentry_out_total_length(child); - /* End-of-directory dentry on disk. */ + /* Account for end-of-directory entry */ *subdir_offset_p += 8; - - /* Recursively call calculate_subdir_offsets() on all the - * children. */ - do { - calculate_subdir_offsets(child, subdir_offset_p); - child = child->next; - } while (child != dentry->inode->children); } else { - /* On disk, childless directories have a valid subdir_offset - * that points to an 8-byte end-of-directory dentry. Regular - * files or reparse points have a subdir_offset of 0. */ - if (dentry_is_directory(dentry)) - *subdir_offset_p += 8; - else - dentry->subdir_offset = 0; + /* Not a directory; set subdir_offset to 0 */ + dentry->subdir_offset = 0; } + return 0; } +/* + * Calculate the subdir offsets for a dentry tree, in preparation of writing + * that dentry tree to a metadata resource. + * + * The subdir offset of each dentry is the offset in the uncompressed metadata + * resource at which its child dentries begin, or 0 if that dentry has no + * children. + * + * The caller must initialize *subdir_offset_p to the first subdir offset that + * is available to use after the root dentry is written. + * + * When this function returns, *subdir_offset_p will have been advanced past the + * size needed for the dentry tree within the uncompressed metadata resource. + */ +void +calculate_subdir_offsets(struct wim_dentry *root, u64 *subdir_offset_p) +{ + for_dentry_in_tree(root, dentry_calculate_subdir_offset, subdir_offset_p); +} -/* Returns the child of @dentry that has the file name @name. - * Returns NULL if no child has the name. */ -struct dentry *get_dentry_child_with_name(const struct dentry *dentry, - const char *name) +/* Compare the UTF-16LE long filenames of two dentries case insensitively. */ +static int +dentry_compare_names_case_insensitive(const struct wim_dentry *d1, + const struct wim_dentry *d2) { - struct dentry *child; - size_t name_len; - - child = dentry->inode->children; - if (child) { - name_len = strlen(name); - do { - if (dentry_has_name(child, name, name_len)) - return child; - child = child->next; - } while (child != dentry->inode->children); - } - return NULL; + return cmp_utf16le_strings(d1->file_name, + d1->file_name_nbytes / 2, + d2->file_name, + d2->file_name_nbytes / 2, + true); } -/* Retrieves the dentry that has the UTF-8 @path relative to the dentry - * @cur_dir. Returns NULL if no dentry having the path is found. */ -static struct dentry *get_dentry_relative_path(struct dentry *cur_dir, - const char *path) +/* Compare the UTF-16LE long filenames of two dentries case sensitively. */ +static int +dentry_compare_names_case_sensitive(const struct wim_dentry *d1, + const struct wim_dentry *d2) { - struct dentry *child; - size_t base_len; - const char *new_path; + return cmp_utf16le_strings(d1->file_name, + d1->file_name_nbytes / 2, + d2->file_name, + d2->file_name_nbytes / 2, + false); +} - if (*path == '\0') - return cur_dir; +static int +_avl_dentry_compare_names_ci(const struct avl_tree_node *n1, + const struct avl_tree_node *n2) +{ + const struct wim_dentry *d1, *d2; - child = cur_dir->inode->children; - if (child) { - new_path = path_next_part(path, &base_len); - do { - if (dentry_has_name(child, path, base_len)) - return get_dentry_relative_path(child, new_path); - child = child->next; - } while (child != cur_dir->inode->children); - } - return NULL; + d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node_ci); + d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node_ci); + return dentry_compare_names_case_insensitive(d1, d2); } -/* Returns the dentry corresponding to the UTF-8 @path, or NULL if there is no - * such dentry. */ -struct dentry *get_dentry(WIMStruct *w, const char *path) +static int +_avl_dentry_compare_names(const struct avl_tree_node *n1, + const struct avl_tree_node *n2) { - struct dentry *root = wim_root_dentry(w); - while (*path == '/') - path++; - return get_dentry_relative_path(root, path); + const struct wim_dentry *d1, *d2; + + d1 = avl_tree_entry(n1, struct wim_dentry, d_index_node); + d2 = avl_tree_entry(n2, struct wim_dentry, d_index_node); + return dentry_compare_names_case_sensitive(d1, d2); } -struct inode *wim_pathname_to_inode(WIMStruct *w, const char *path) +/* Default case sensitivity behavior for searches with + * WIMLIB_CASE_PLATFORM_DEFAULT specified. This can be modified by passing + * WIMLIB_INIT_FLAG_DEFAULT_CASE_SENSITIVE or + * WIMLIB_INIT_FLAG_DEFAULT_CASE_INSENSITIVE to wimlib_global_init(). */ +bool default_ignore_case = +#ifdef __WIN32__ + true +#else + false +#endif +; + +/* Case-sensitive dentry lookup. Only @file_name and @file_name_nbytes of + * @dummy must be valid. */ +static struct wim_dentry * +dir_lookup(const struct wim_inode *dir, const struct wim_dentry *dummy) { - struct dentry *dentry; - dentry = get_dentry(w, path); - if (!dentry) + struct avl_tree_node *node; + + node = avl_tree_lookup_node(dir->i_children, + &dummy->d_index_node, + _avl_dentry_compare_names); + if (!node) return NULL; - else - return dentry->inode; + return avl_tree_entry(node, struct wim_dentry, d_index_node); } -/* Returns the dentry that corresponds to the parent directory of @path, or NULL - * if the dentry is not found. */ -struct dentry *get_parent_dentry(WIMStruct *w, const char *path) +/* Case-insensitive dentry lookup. Only @file_name and @file_name_nbytes of + * @dummy must be valid. */ +static struct wim_dentry * +dir_lookup_ci(const struct wim_inode *dir, const struct wim_dentry *dummy) { - size_t path_len = strlen(path); - char buf[path_len + 1]; + struct avl_tree_node *node; + + node = avl_tree_lookup_node(dir->i_children_ci, + &dummy->d_index_node_ci, + _avl_dentry_compare_names_ci); + if (!node) + return NULL; + return avl_tree_entry(node, struct wim_dentry, d_index_node_ci); +} - memcpy(buf, path, path_len + 1); +/* Given a UTF-16LE filename and a directory, look up the dentry for the file. + * Return it if found, otherwise NULL. This has configurable case sensitivity, + * and @name need not be null-terminated. */ +struct wim_dentry * +get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry, + const utf16lechar *name, + size_t name_nbytes, + CASE_SENSITIVITY_TYPE case_ctype) +{ + const struct wim_inode *dir = dentry->d_inode; + bool ignore_case = will_ignore_case(case_ctype); + struct wim_dentry dummy; + struct wim_dentry *child; - to_parent_name(buf, path_len); + dummy.file_name = (utf16lechar*)name; + dummy.file_name_nbytes = name_nbytes; + + if (!ignore_case) + /* Case-sensitive lookup. */ + return dir_lookup(dir, &dummy); + + /* Case-insensitive lookup. */ + + child = dir_lookup_ci(dir, &dummy); + if (!child) + return NULL; + + if (likely(list_empty(&child->d_ci_conflict_list))) + /* Only one dentry has this case-insensitive name; return it */ + return child; + + /* Multiple dentries have the same case-insensitive name. Choose the + * dentry with the same case-sensitive name, if one exists; otherwise + * print a warning and choose one of the possible dentries arbitrarily. + */ + struct wim_dentry *alt = child; + size_t num_alts = 0; - return get_dentry(w, buf); + do { + num_alts++; + if (!dentry_compare_names_case_sensitive(&dummy, alt)) + return alt; + alt = list_entry(alt->d_ci_conflict_list.next, + struct wim_dentry, d_ci_conflict_list); + } while (alt != child); + + WARNING("Result of case-insensitive lookup is ambiguous\n" + " (returning \"%"TS"\" of %zu " + "possible files, including \"%"TS"\")", + dentry_full_path(child), + num_alts, + dentry_full_path(list_entry(child->d_ci_conflict_list.next, + struct wim_dentry, + d_ci_conflict_list))); + return child; } -/* Prints the full path of a dentry. */ -int print_dentry_full_path(struct dentry *dentry, void *ignore) +/* Given a 'tchar' filename and a directory, look up the dentry for the file. + * If the filename was successfully converted to UTF-16LE and the dentry was + * found, return it; otherwise return NULL. This has configurable case + * sensitivity. */ +struct wim_dentry * +get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name, + CASE_SENSITIVITY_TYPE case_type) { - if (dentry->full_path_utf8) - puts(dentry->full_path_utf8); - return 0; + int ret; + const utf16lechar *name_utf16le; + size_t name_utf16le_nbytes; + struct wim_dentry *child; + + ret = tstr_get_utf16le_and_len(name, &name_utf16le, + &name_utf16le_nbytes); + if (ret) + return NULL; + + child = get_dentry_child_with_utf16le_name(dentry, + name_utf16le, + name_utf16le_nbytes, + case_type); + tstr_put_utf16le(name_utf16le); + return child; } -/* We want to be able to show the names of the file attribute flags that are - * set. */ -struct file_attr_flag { - u32 flag; - const char *name; -}; -struct file_attr_flag file_attr_flags[] = { - {FILE_ATTRIBUTE_READONLY, "READONLY"}, - {FILE_ATTRIBUTE_HIDDEN, "HIDDEN"}, - {FILE_ATTRIBUTE_SYSTEM, "SYSTEM"}, - {FILE_ATTRIBUTE_DIRECTORY, "DIRECTORY"}, - {FILE_ATTRIBUTE_ARCHIVE, "ARCHIVE"}, - {FILE_ATTRIBUTE_DEVICE, "DEVICE"}, - {FILE_ATTRIBUTE_NORMAL, "NORMAL"}, - {FILE_ATTRIBUTE_TEMPORARY, "TEMPORARY"}, - {FILE_ATTRIBUTE_SPARSE_FILE, "SPARSE_FILE"}, - {FILE_ATTRIBUTE_REPARSE_POINT, "REPARSE_POINT"}, - {FILE_ATTRIBUTE_COMPRESSED, "COMPRESSED"}, - {FILE_ATTRIBUTE_OFFLINE, "OFFLINE"}, - {FILE_ATTRIBUTE_NOT_CONTENT_INDEXED,"NOT_CONTENT_INDEXED"}, - {FILE_ATTRIBUTE_ENCRYPTED, "ENCRYPTED"}, - {FILE_ATTRIBUTE_VIRTUAL, "VIRTUAL"}, -}; - -/* Prints a directory entry. @lookup_table is a pointer to the lookup table, if - * available. If the dentry is unresolved and the lookup table is NULL, the - * lookup table entries will not be printed. Otherwise, they will be. */ -int print_dentry(struct dentry *dentry, void *lookup_table) +/* This is the UTF-16LE version of get_dentry(), currently private to this file + * because no one needs it besides get_dentry(). */ +static struct wim_dentry * +get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path, + CASE_SENSITIVITY_TYPE case_type) { - const u8 *hash; - struct lookup_table_entry *lte; - const struct inode *inode = dentry->inode; - time_t time; - char *p; - - printf("[DENTRY]\n"); - printf("Length = %"PRIu64"\n", dentry->length); - printf("Attributes = 0x%x\n", inode->attributes); - for (unsigned i = 0; i < ARRAY_LEN(file_attr_flags); i++) - if (file_attr_flags[i].flag & inode->attributes) - printf(" FILE_ATTRIBUTE_%s is set\n", - file_attr_flags[i].name); - printf("Security ID = %d\n", inode->security_id); - printf("Subdir offset = %"PRIu64"\n", dentry->subdir_offset); - - /* Translate the timestamps into something readable */ - time = wim_timestamp_to_unix(inode->creation_time); - p = asctime(gmtime(&time)); - *(strrchr(p, '\n')) = '\0'; - printf("Creation Time = %s UTC\n", p); - - time = wim_timestamp_to_unix(inode->last_access_time); - p = asctime(gmtime(&time)); - *(strrchr(p, '\n')) = '\0'; - printf("Last Access Time = %s UTC\n", p); - - time = wim_timestamp_to_unix(inode->last_write_time); - p = asctime(gmtime(&time)); - *(strrchr(p, '\n')) = '\0'; - printf("Last Write Time = %s UTC\n", p); - - printf("Reparse Tag = 0x%"PRIx32"\n", inode->reparse_tag); - printf("Hard Link Group = 0x%"PRIx64"\n", inode->ino); - printf("Hard Link Group Size = %"PRIu32"\n", inode->link_count); - printf("Number of Alternate Data Streams = %hu\n", inode->num_ads); - printf("Filename = \""); - print_string(dentry->file_name, dentry->file_name_len); - puts("\""); - printf("Filename Length = %hu\n", dentry->file_name_len); - printf("Filename (UTF-8) = \"%s\"\n", dentry->file_name_utf8); - printf("Filename (UTF-8) Length = %hu\n", dentry->file_name_utf8_len); - printf("Short Name = \""); - print_string(dentry->short_name, dentry->short_name_len); - puts("\""); - printf("Short Name Length = %hu\n", dentry->short_name_len); - printf("Full Path (UTF-8) = \"%s\"\n", dentry->full_path_utf8); - lte = inode_stream_lte(dentry->inode, 0, lookup_table); - if (lte) { - print_lookup_table_entry(lte); - } else { - hash = inode_stream_hash(inode, 0); - if (hash) { - printf("Hash = 0x"); - print_hash(hash); - putchar('\n'); - putchar('\n'); + struct wim_dentry *cur_dentry; + const utf16lechar *name_start, *name_end; + + /* Start with the root directory of the image. Note: this will be NULL + * if an image has been added directly with wimlib_add_empty_image() but + * no files have been added yet; in that case we fail with ENOENT. */ + cur_dentry = wim_get_current_root_dentry(wim); + + name_start = path; + for (;;) { + if (cur_dentry == NULL) { + errno = ENOENT; + return NULL; } - } - for (u16 i = 0; i < inode->num_ads; i++) { - printf("[Alternate Stream Entry %u]\n", i); - printf("Name = \"%s\"\n", inode->ads_entries[i]->stream_name_utf8); - printf("Name Length (UTF-16) = %u\n", - inode->ads_entries[i]->stream_name_len); - hash = inode_stream_hash(inode, i + 1); - if (hash) { - printf("Hash = 0x"); - print_hash(hash); - putchar('\n'); + + if (*name_start && !dentry_is_directory(cur_dentry)) { + errno = ENOTDIR; + return NULL; } - print_lookup_table_entry(inode_stream_lte(inode, i + 1, - lookup_table)); + + while (*name_start == cpu_to_le16(WIM_PATH_SEPARATOR)) + name_start++; + + if (!*name_start) + return cur_dentry; + + name_end = name_start; + do { + ++name_end; + } while (*name_end != cpu_to_le16(WIM_PATH_SEPARATOR) && *name_end); + + cur_dentry = get_dentry_child_with_utf16le_name(cur_dentry, + name_start, + (u8*)name_end - (u8*)name_start, + case_type); + name_start = name_end; } - return 0; } -/* Initializations done on every `struct dentry'. */ -static void dentry_common_init(struct dentry *dentry) +/* + * WIM path lookup: translate a path in the currently selected WIM image to the + * corresponding dentry, if it exists. + * + * @wim + * The WIMStruct for the WIM. The search takes place in the currently + * selected image. + * + * @path + * The path to look up, given relative to the root of the WIM image. + * Characters with value WIM_PATH_SEPARATOR are taken to be path + * separators. Leading path separators are ignored, whereas one or more + * trailing path separators cause the path to only match a directory. + * + * @case_type + * The case-sensitivity behavior of this function, as one of the following + * constants: + * + * - WIMLIB_CASE_SENSITIVE: Perform the search case sensitively. This means + * that names must match exactly. + * + * - WIMLIB_CASE_INSENSITIVE: Perform the search case insensitively. This + * means that names are considered to match if they are equal when + * transformed to upper case. If a path component matches multiple names + * case-insensitively, the name that matches the path component + * case-sensitively is chosen, if existent; otherwise one + * case-insensitively matching name is chosen arbitrarily. + * + * - WIMLIB_CASE_PLATFORM_DEFAULT: Perform either case-sensitive or + * case-insensitive search, depending on the value of the global variable + * default_ignore_case. + * + * In any case, no Unicode normalization is done before comparing strings. + * + * Returns a pointer to the dentry that is the result of the lookup, or NULL if + * no such dentry exists. If NULL is returned, errno is set to one of the + * following values: + * + * ENOTDIR if one of the path components used as a directory existed but + * was not, in fact, a directory. + * + * ENOENT otherwise. + * + * Additional notes: + * + * - This function does not consider a reparse point to be a directory, even + * if it has FILE_ATTRIBUTE_DIRECTORY set. + * + * - This function does not dereference symbolic links or junction points + * when performing the search. + * + * - Since this function ignores leading slashes, the empty path is valid and + * names the root directory of the WIM image. + * + * - An image added with wimlib_add_empty_image() does not have a root + * directory yet, and this function will fail with ENOENT for any path on + * such an image. + */ +struct wim_dentry * +get_dentry(WIMStruct *wim, const tchar *path, CASE_SENSITIVITY_TYPE case_type) { - memset(dentry, 0, sizeof(struct dentry)); - dentry->refcnt = 1; + int ret; + const utf16lechar *path_utf16le; + struct wim_dentry *dentry; + + ret = tstr_get_utf16le(path, &path_utf16le); + if (ret) + return NULL; + dentry = get_dentry_utf16le(wim, path_utf16le, case_type); + tstr_put_utf16le(path_utf16le); + return dentry; } -struct inode *new_timeless_inode() +/* Modify @path, which is a null-terminated string @len 'tchars' in length, + * in-place to produce the path to its parent directory. */ +static void +to_parent_name(tchar *path, size_t len) { - struct inode *inode = CALLOC(1, sizeof(struct inode)); - if (!inode) - return NULL; - inode->security_id = -1; - inode->link_count = 1; - INIT_LIST_HEAD(&inode->dentry_list); - return inode; + ssize_t i = (ssize_t)len - 1; + while (i >= 0 && path[i] == WIM_PATH_SEPARATOR) + i--; + while (i >= 0 && path[i] != WIM_PATH_SEPARATOR) + i--; + while (i >= 0 && path[i] == WIM_PATH_SEPARATOR) + i--; + path[i + 1] = T('\0'); } -struct inode *new_inode() +/* Similar to get_dentry(), but returns the dentry named by @path with the last + * component stripped off. + * + * Note: The returned dentry is NOT guaranteed to be a directory. */ +struct wim_dentry * +get_parent_dentry(WIMStruct *wim, const tchar *path, + CASE_SENSITIVITY_TYPE case_type) { - struct inode *inode = new_timeless_inode(); - if (!inode) - return NULL; - u64 now = get_wim_timestamp(); - inode->creation_time = now; - inode->last_access_time = now; - inode->last_write_time = now; - return inode; + size_t path_len = tstrlen(path); + tchar buf[path_len + 1]; + + tmemcpy(buf, path, path_len + 1); + to_parent_name(buf, path_len); + return get_dentry(wim, buf, case_type); } -/* - * Creates an unlinked directory entry. +/* + * Create an unlinked dentry. * - * @name: The UTF-8 filename of the new dentry. + * @name specifies the long name to give the new dentry. If NULL or empty, the + * new dentry will be given no long name. * - * Returns a pointer to the new dentry, or NULL if out of memory. + * The new dentry will have no short name and no associated inode. + * + * On success, returns 0 and a pointer to the new, allocated dentry is stored in + * *dentry_ret. On failure, returns WIMLIB_ERR_NOMEM or an error code resulting + * from a failed string conversion. */ -struct dentry *new_dentry(const char *name) +static int +new_dentry(const tchar *name, struct wim_dentry **dentry_ret) { - struct dentry *dentry; - - dentry = MALLOC(sizeof(struct dentry)); - if (!dentry) - goto err; - - dentry_common_init(dentry); - if (change_dentry_name(dentry, name) != 0) - goto err; + struct wim_dentry *dentry; + int ret; - dentry->next = dentry; - dentry->prev = dentry; - dentry->parent = dentry; + dentry = CALLOC(1, sizeof(struct wim_dentry)); + if (!dentry) + return WIMLIB_ERR_NOMEM; - return dentry; -err: - FREE(dentry); - ERROR("Failed to allocate new dentry"); - return NULL; + if (name && *name) { + ret = dentry_set_name(dentry, name); + if (ret) { + FREE(dentry); + return ret; + } + } + dentry->d_parent = dentry; + *dentry_ret = dentry; + return 0; } -struct dentry *new_dentry_with_inode(const char *name) +/* Like new_dentry(), but also allocate an inode and associate it with the + * dentry. If set_timestamps=true, the timestamps for the inode will be set to + * the current time; otherwise, they will be left 0. */ +int +new_dentry_with_new_inode(const tchar *name, bool set_timestamps, + struct wim_dentry **dentry_ret) { - struct dentry *dentry; - dentry = new_dentry(name); - if (dentry) { - dentry->inode = new_inode(); - if (dentry->inode) { - inode_add_dentry(dentry, dentry->inode); - } else { - free_dentry(dentry); - dentry = NULL; - } + struct wim_dentry *dentry; + struct wim_inode *inode; + int ret; + + ret = new_dentry(name, &dentry); + if (ret) + return ret; + + inode = new_inode(dentry, set_timestamps); + if (!inode) { + free_dentry(dentry); + return WIMLIB_ERR_NOMEM; } - return dentry; + + *dentry_ret = dentry; + return 0; } -void free_ads_entry(struct ads_entry *entry) +/* Like new_dentry(), but also associate the new dentry with the specified inode + * and acquire a reference to each of the inode's blobs. */ +int +new_dentry_with_existing_inode(const tchar *name, struct wim_inode *inode, + struct wim_dentry **dentry_ret) { - if (entry) { - FREE(entry->stream_name); - FREE(entry->stream_name_utf8); - FREE(entry); - } + int ret = new_dentry(name, dentry_ret); + if (ret) + return ret; + d_associate(*dentry_ret, inode); + inode_ref_blobs(inode); + return 0; } -void inode_free_ads_entries(struct inode *inode) +/* Create an unnamed dentry with a new inode for a directory with the default + * metadata. */ +int +new_filler_directory(struct wim_dentry **dentry_ret) { - if (inode->ads_entries) { - for (u16 i = 0; i < inode->num_ads; i++) - free_ads_entry(inode->ads_entries[i]); - FREE(inode->ads_entries); - } + int ret; + struct wim_dentry *dentry; + + ret = new_dentry_with_new_inode(NULL, true, &dentry); + if (ret) + return ret; + /* Leave the inode number as 0; this is allowed for non + * hard-linked files. */ + dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY; + *dentry_ret = dentry; + return 0; } -/* Frees an inode. */ -void free_inode(struct inode *inode) +static int +dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore) { - if (inode) { - inode_free_ads_entries(inode); - #ifdef WITH_FUSE - wimlib_assert(inode->num_opened_fds == 0); - FREE(inode->fds); - #endif - FREE(inode); - } + dentry->d_inode->i_visited = 0; + return 0; } -/* Decrements link count on an inode and frees it if the link count reaches 0. - * */ -struct inode *put_inode(struct inode *inode) +void +dentry_tree_clear_inode_visited(struct wim_dentry *root) { - if (inode) { - wimlib_assert(inode->link_count); - if (--inode->link_count == 0) { - #ifdef WITH_FUSE - if (inode->num_opened_fds == 0) - #endif - { - free_inode(inode); - inode = NULL; - } - } - } - return inode; + for_dentry_in_tree(root, dentry_clear_inode_visited, NULL); } -/* Frees a WIM dentry. +/* + * Free a WIM dentry. * - * The inode is freed only if its link count is decremented to 0. + * In addition to freeing the dentry itself, this disassociates the dentry from + * its inode. If the inode is no longer in use, it will be freed as well. */ -struct inode *free_dentry(struct dentry *dentry) +void +free_dentry(struct wim_dentry *dentry) { - wimlib_assert(dentry); - struct inode *inode; - - FREE(dentry->file_name); - FREE(dentry->file_name_utf8); - FREE(dentry->short_name); - FREE(dentry->full_path_utf8); - inode = put_inode(dentry->inode); - FREE(dentry); - return inode; + if (dentry) { + d_disassociate(dentry); + FREE(dentry->file_name); + FREE(dentry->short_name); + FREE(dentry->_full_path); + FREE(dentry); + } } -void put_dentry(struct dentry *dentry) +static int +do_free_dentry(struct wim_dentry *dentry, void *_ignore) { - wimlib_assert(dentry); - wimlib_assert(dentry->refcnt); + free_dentry(dentry); + return 0; +} - if (--dentry->refcnt == 0) - free_dentry(dentry); +static int +do_free_dentry_and_unref_blobs(struct wim_dentry *dentry, void *blob_table) +{ + inode_unref_blobs(dentry->d_inode, blob_table); + free_dentry(dentry); + return 0; } -#if 0 -/* Partically clones a dentry. +/* + * Free all dentries in a tree. + * + * @root: + * The root of the dentry tree to free. If NULL, this function has no + * effect. + * + * @blob_table: + * A pointer to the blob table for the WIM, or NULL if not specified. If + * specified, this function will decrement the reference counts of the + * blobs referenced by the dentries. * - * Beware: - * - memory for file names is not cloned (the pointers are all set to NULL - * and the lengths are set to zero) - * - next, prev, and children pointers and not touched + * This function also releases references to the corresponding inodes. + * + * This function does *not* unlink @root from its parent directory, if it has + * one. If @root has a parent, the caller must unlink @root before calling this + * function. */ -struct dentry *clone_dentry(struct dentry *old) +void +free_dentry_tree(struct wim_dentry *root, struct blob_table *blob_table) { - struct dentry *new = MALLOC(sizeof(struct dentry)); - if (!new) - return NULL; - memcpy(new, old, sizeof(struct dentry)); - new->file_name = NULL; - new->file_name_len = 0; - new->file_name_utf8 = NULL; - new->file_name_utf8_len = 0; - new->short_name = NULL; - new->short_name_len = 0; - return new; + int (*f)(struct wim_dentry *, void *); + + if (blob_table) + f = do_free_dentry_and_unref_blobs; + else + f = do_free_dentry; + + for_dentry_in_tree_depth(root, f, blob_table); } -#endif -/* - * This function is passed as an argument to for_dentry_in_tree_depth() in order - * to free a directory tree. __args is a pointer to a `struct free_dentry_args'. - */ -static int do_free_dentry(struct dentry *dentry, void *__lookup_table) +/* Insert the @child dentry into the case sensitive index of the @dir directory. + * Return NULL if successfully inserted, otherwise a pointer to the + * already-inserted duplicate. */ +static struct wim_dentry * +dir_index_child(struct wim_inode *dir, struct wim_dentry *child) { - struct lookup_table *lookup_table = __lookup_table; - struct lookup_table_entry *lte; - struct inode *inode = dentry->inode; - unsigned i; - - if (lookup_table) { - wimlib_assert(inode->link_count); - for (i = 0; i <= inode->num_ads; i++) { - lte = inode_stream_lte(inode, i, lookup_table); - lte_decrement_refcnt(lte, lookup_table); - } - } + struct avl_tree_node *duplicate; - wimlib_assert(dentry->refcnt != 0); - if (--dentry->refcnt == 0) - free_dentry(dentry); - return 0; + duplicate = avl_tree_insert(&dir->i_children, + &child->d_index_node, + _avl_dentry_compare_names); + if (!duplicate) + return NULL; + return avl_tree_entry(duplicate, struct wim_dentry, d_index_node); } -/* - * Unlinks and frees a dentry tree. - * - * @root: The root of the tree. - * @lookup_table: The lookup table for dentries. If non-NULL, the - * reference counts in the lookup table for the lookup - * table entries corresponding to the dentries will be - * decremented. - */ -void free_dentry_tree(struct dentry *root, struct lookup_table *lookup_table) +/* Insert the @child dentry into the case insensitive index of the @dir + * directory. Return NULL if successfully inserted, otherwise a pointer to the + * already-inserted duplicate. */ +static struct wim_dentry * +dir_index_child_ci(struct wim_inode *dir, struct wim_dentry *child) { - if (!root || !root->parent) - return; - for_dentry_in_tree_depth(root, do_free_dentry, lookup_table); + struct avl_tree_node *duplicate; + + duplicate = avl_tree_insert(&dir->i_children_ci, + &child->d_index_node_ci, + _avl_dentry_compare_names_ci); + if (!duplicate) + return NULL; + return avl_tree_entry(duplicate, struct wim_dentry, d_index_node_ci); } -int increment_dentry_refcnt(struct dentry *dentry, void *ignore) +/* Remove the specified dentry from its directory's case-sensitive index. */ +static void +dir_unindex_child(struct wim_inode *dir, struct wim_dentry *child) { - dentry->refcnt++; - return 0; + avl_tree_remove(&dir->i_children, &child->d_index_node); } -/* - * Links a dentry into the directory tree. - * - * @dentry: The dentry to link. - * @parent: The dentry that will be the parent of @dentry. - */ -void link_dentry(struct dentry *dentry, struct dentry *parent) +/* Remove the specified dentry from its directory's case-insensitive index. */ +static void +dir_unindex_child_ci(struct wim_inode *dir, struct wim_dentry *child) { - wimlib_assert(dentry_is_directory(parent)); - dentry->parent = parent; - if (parent->inode->children) { - /* Not an only child; link to siblings. */ - dentry->next = parent->inode->children; - dentry->prev = parent->inode->children->prev; - dentry->next->prev = dentry; - dentry->prev->next = dentry; - } else { - /* Only child; link to parent. */ - parent->inode->children = dentry; - dentry->next = dentry; - dentry->prev = dentry; - } + avl_tree_remove(&dir->i_children_ci, &child->d_index_node_ci); } +/* Return true iff the specified dentry is in its parent directory's + * case-insensitive index. */ +static bool +dentry_in_ci_index(const struct wim_dentry *dentry) +{ + return !avl_tree_node_is_unlinked(&dentry->d_index_node_ci); +} -/* - * Unlink a dentry from the directory tree. +/* + * Link a dentry into the tree. + * + * @parent: + * The dentry that will be the parent of @child. It must name a directory. * - * Note: This merely removes it from the in-memory tree structure. + * @child: + * The dentry to link. It must be currently unlinked. + * + * Returns NULL if successful. If @parent already contains a dentry with the + * same case-sensitive name as @child, returns a pointer to this duplicate + * dentry. */ -void unlink_dentry(struct dentry *dentry) +struct wim_dentry * +dentry_add_child(struct wim_dentry *parent, struct wim_dentry *child) { - if (dentry_is_root(dentry)) - return; - if (dentry_is_only_child(dentry)) { - dentry->parent->inode->children = NULL; + struct wim_dentry *duplicate; + struct wim_inode *dir; + + wimlib_assert(parent != child); + + dir = parent->d_inode; + + wimlib_assert(inode_is_directory(dir)); + + duplicate = dir_index_child(dir, child); + if (duplicate) + return duplicate; + + duplicate = dir_index_child_ci(dir, child); + if (duplicate) { + list_add(&child->d_ci_conflict_list, &duplicate->d_ci_conflict_list); + avl_tree_node_set_unlinked(&child->d_index_node_ci); } else { - if (dentry_is_first_sibling(dentry)) - dentry->parent->inode->children = dentry->next; - dentry->next->prev = dentry->prev; - dentry->prev->next = dentry->next; + INIT_LIST_HEAD(&child->d_ci_conflict_list); } + child->d_parent = parent; + return NULL; } -/* Duplicates a UTF-8 name into UTF-8 and UTF-16 strings and returns the strings - * and their lengths in the pointer arguments */ -int get_names(char **name_utf16_ret, char **name_utf8_ret, - u16 *name_utf16_len_ret, u16 *name_utf8_len_ret, - const char *name) +/* Unlink a dentry from the tree. */ +void +unlink_dentry(struct wim_dentry *dentry) { - size_t utf8_len; - size_t utf16_len; - char *name_utf16, *name_utf8; + struct wim_inode *dir; - utf8_len = strlen(name); + /* Do nothing if the dentry is root or it's already unlinked. Not + * actually necessary based on the current callers, but we do the check + * here to be safe. */ + if (unlikely(dentry->d_parent == dentry)) + return; - name_utf16 = utf8_to_utf16(name, utf8_len, &utf16_len); + dir = dentry->d_parent->d_inode; - if (!name_utf16) - return WIMLIB_ERR_NOMEM; + dir_unindex_child(dir, dentry); - name_utf8 = MALLOC(utf8_len + 1); - if (!name_utf8) { - FREE(name_utf8); - return WIMLIB_ERR_NOMEM; + if (dentry_in_ci_index(dentry)) { + + dir_unindex_child_ci(dir, dentry); + + if (!list_empty(&dentry->d_ci_conflict_list)) { + /* Make a different case-insensitively-the-same dentry + * be the "representative" in the search index. */ + struct list_head *next; + struct wim_dentry *other; + struct wim_dentry *existing; + + next = dentry->d_ci_conflict_list.next; + other = list_entry(next, struct wim_dentry, d_ci_conflict_list); + existing = dir_index_child_ci(dir, other); + wimlib_assert(existing == NULL); + } } - memcpy(name_utf8, name, utf8_len + 1); - FREE(*name_utf8_ret); - FREE(*name_utf16_ret); - *name_utf8_ret = name_utf8; - *name_utf16_ret = name_utf16; - *name_utf8_len_ret = utf8_len; - *name_utf16_len_ret = utf16_len; - return 0; + list_del(&dentry->d_ci_conflict_list); + + /* Not actually necessary, but to be safe don't retain the now-obsolete + * parent pointer. */ + dentry->d_parent = dentry; } -/* Changes the name of a dentry to @new_name. Only changes the file_name and - * file_name_utf8 fields; does not change the short_name, short_name_utf8, or - * full_path_utf8 fields. Also recalculates its length. */ -int change_dentry_name(struct dentry *dentry, const char *new_name) +static int +read_extra_data(const u8 *p, const u8 *end, struct wim_inode *inode) { - int ret; + while (((uintptr_t)p & 7) && p < end) + p++; - ret = get_names(&dentry->file_name, &dentry->file_name_utf8, - &dentry->file_name_len, &dentry->file_name_utf8_len, - new_name); - FREE(dentry->short_name); - dentry->short_name_len = 0; - if (ret == 0) - dentry->length = dentry_correct_length(dentry); - return ret; + if (unlikely(p < end)) { + inode->i_extra = memdup(p, end - p); + if (!inode->i_extra) + return WIMLIB_ERR_NOMEM; + inode->i_extra_size = end - p; + } + return 0; } /* - * Changes the name of an alternate data stream */ -int change_ads_name(struct ads_entry *entry, const char *new_name) + * Set the type of each stream for an encrypted file. + * + * All data streams of the encrypted file should have been packed into a single + * stream in the format provided by ReadEncryptedFileRaw() on Windows. We + * assign this stream type STREAM_TYPE_EFSRPC_RAW_DATA. + * + * Encrypted files can't have a reparse point stream. In the on-disk NTFS + * format they can, but as far as I know the reparse point stream of an + * encrypted file can't be stored in the WIM format in a way that's compatible + * with WIMGAPI, nor is there even any way for it to be read or written on + * Windows when the process does not have access to the file encryption key. + */ +static void +assign_stream_types_encrypted(struct wim_inode *inode) { - return get_names(&entry->stream_name, &entry->stream_name_utf8, - &entry->stream_name_len, - &entry->stream_name_utf8_len, - new_name); + for (unsigned i = 0; i < inode->i_num_streams; i++) { + struct wim_inode_stream *strm = &inode->i_streams[i]; + if (!stream_is_named(strm) && !is_zero_hash(strm->_stream_hash)) + { + strm->stream_type = STREAM_TYPE_EFSRPC_RAW_DATA; + return; + } + } } -/* Parameters for calculate_dentry_statistics(). */ -struct image_statistics { - struct lookup_table *lookup_table; - u64 *dir_count; - u64 *file_count; - u64 *total_bytes; - u64 *hard_link_bytes; -}; - -static int calculate_dentry_statistics(struct dentry *dentry, void *arg) +/* + * Set the type of each stream for an unencrypted file. + * + * There will be an unnamed data stream, a reparse point stream, or both an + * unnamed data stream and a reparse point stream. In addition, there may be + * named data streams. + */ +static void +assign_stream_types_unencrypted(struct wim_inode *inode) { - struct image_statistics *stats; - struct lookup_table_entry *lte; - - stats = arg; - - if (dentry_is_directory(dentry) && !dentry_is_root(dentry)) - ++*stats->dir_count; - else - ++*stats->file_count; - - for (unsigned i = 0; i <= dentry->inode->num_ads; i++) { - lte = inode_stream_lte(dentry->inode, i, stats->lookup_table); - if (lte) { - *stats->total_bytes += wim_resource_size(lte); - if (++lte->out_refcnt == 1) - *stats->hard_link_bytes += wim_resource_size(lte); + bool found_reparse_point_stream = false; + bool found_unnamed_data_stream = false; + struct wim_inode_stream *unnamed_stream_with_zero_hash = NULL; + + for (unsigned i = 0; i < inode->i_num_streams; i++) { + struct wim_inode_stream *strm = &inode->i_streams[i]; + + if (stream_is_named(strm)) { + /* Named data stream */ + strm->stream_type = STREAM_TYPE_DATA; + } else if (!is_zero_hash(strm->_stream_hash)) { + if ((inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) && + !found_reparse_point_stream) { + found_reparse_point_stream = true; + strm->stream_type = STREAM_TYPE_REPARSE_POINT; + } else if (!found_unnamed_data_stream) { + found_unnamed_data_stream = true; + strm->stream_type = STREAM_TYPE_DATA; + } + } else { + /* If no stream name is specified and the hash is zero, + * then remember this stream for later so that we can + * assign it to the unnamed data stream if we don't find + * a better candidate. */ + unnamed_stream_with_zero_hash = strm; } } - return 0; -} -/* Calculates some statistics about a dentry tree. */ -void calculate_dir_tree_statistics(struct dentry *root, struct lookup_table *table, - u64 *dir_count_ret, u64 *file_count_ret, - u64 *total_bytes_ret, - u64 *hard_link_bytes_ret) -{ - struct image_statistics stats; - *dir_count_ret = 0; - *file_count_ret = 0; - *total_bytes_ret = 0; - *hard_link_bytes_ret = 0; - stats.lookup_table = table; - stats.dir_count = dir_count_ret; - stats.file_count = file_count_ret; - stats.total_bytes = total_bytes_ret; - stats.hard_link_bytes = hard_link_bytes_ret; - for_lookup_table_entry(table, zero_out_refcnts, NULL); - for_dentry_in_tree(root, calculate_dentry_statistics, &stats); + if (!found_unnamed_data_stream && unnamed_stream_with_zero_hash != NULL) + unnamed_stream_with_zero_hash->stream_type = STREAM_TYPE_DATA; } - -/* - * Reads the alternate data stream entries for a dentry. - * - * @p: Pointer to buffer that starts with the first alternate stream entry. - * - * @inode: Inode to load the alternate data streams into. - * @inode->num_ads must have been set to the number of - * alternate data streams that are expected. - * - * @remaining_size: Number of bytes of data remaining in the buffer pointed - * to by @p. - * - * The format of the on-disk alternate stream entries is as follows: - * - * struct ads_entry_on_disk { - * u64 length; // Length of the entry, in bytes. This includes - * all fields (including the stream name and - * null terminator if present, AND the padding!). - * u64 reserved; // Seems to be unused - * u8 hash[20]; // SHA1 message digest of the uncompressed stream - * u16 stream_name_len; // Length of the stream name, in bytes - * char stream_name[]; // Stream name in UTF-16LE, @stream_name_len bytes long, - * not including null terminator - * u16 zero; // UTF-16 null terminator for the stream name, NOT - * included in @stream_name_len. Based on what - * I've observed from filenames in dentries, - * this field should not exist when - * (@stream_name_len == 0), but you can't - * actually tell because of the padding anyway - * (provided that the padding is zeroed, which - * it always seems to be). - * char padding[]; // Padding to make the size a multiple of 8 bytes. - * }; - * - * In addition, the entries are 8-byte aligned. - * - * Return 0 on success or nonzero on failure. On success, inode->ads_entries - * is set to an array of `struct ads_entry's of length inode->num_ads. On - * failure, @inode is not modified. +/* + * Read and interpret the collection of streams for the specified inode. */ -static int read_ads_entries(const u8 *p, struct inode *inode, - u64 remaining_size) +static int +setup_inode_streams(const u8 *p, const u8 *end, struct wim_inode *inode, + unsigned num_extra_streams, const u8 *default_hash, + u64 *offset_p) { - u16 num_ads; - struct ads_entry **ads_entries; - int ret; + const u8 *orig_p = p; - num_ads = inode->num_ads; - ads_entries = CALLOC(num_ads, sizeof(inode->ads_entries[0])); - if (!ads_entries) { - ERROR("Could not allocate memory for %"PRIu16" " - "alternate data stream entries", num_ads); - return WIMLIB_ERR_NOMEM; + inode->i_num_streams = 1 + num_extra_streams; + + if (unlikely(inode->i_num_streams > ARRAY_LEN(inode->i_embedded_streams))) { + inode->i_streams = CALLOC(inode->i_num_streams, + sizeof(inode->i_streams[0])); + if (!inode->i_streams) + return WIMLIB_ERR_NOMEM; } - for (u16 i = 0; i < num_ads; i++) { - struct ads_entry *cur_entry; + /* Use the default hash field for the first stream */ + inode->i_streams[0].stream_name = (utf16lechar *)NO_STREAM_NAME; + copy_hash(inode->i_streams[0]._stream_hash, default_hash); + inode->i_streams[0].stream_type = STREAM_TYPE_UNKNOWN; + inode->i_streams[0].stream_id = 0; + + /* Read the extra stream entries */ + for (unsigned i = 1; i < inode->i_num_streams; i++) { + struct wim_inode_stream *strm; + const struct wim_extra_stream_entry_on_disk *disk_strm; u64 length; - u64 length_no_padding; - u64 total_length; - size_t utf8_len; - const u8 *p_save = p; + u16 name_nbytes; - cur_entry = new_ads_entry(NULL); - if (!cur_entry) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ads_entries; - } + strm = &inode->i_streams[i]; - ads_entries[i] = cur_entry; + strm->stream_id = i; - /* Read the base stream entry, excluding the stream name. */ - if (remaining_size < WIM_ADS_ENTRY_DISK_SIZE) { - ERROR("Stream entries go past end of metadata resource"); - ERROR("(remaining_size = %"PRIu64")", remaining_size); - ret = WIMLIB_ERR_INVALID_DENTRY; - goto out_free_ads_entries; - } + /* Do we have at least the size of the fixed-length data we know + * need? */ + if ((end - p) < sizeof(struct wim_extra_stream_entry_on_disk)) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - p = get_u64(p, &length); - p += 8; /* Skip the reserved field */ - p = get_bytes(p, SHA1_HASH_SIZE, (u8*)cur_entry->hash); - p = get_u16(p, &cur_entry->stream_name_len); + disk_strm = (const struct wim_extra_stream_entry_on_disk *)p; - cur_entry->stream_name = NULL; - cur_entry->stream_name_utf8 = NULL; + /* Read the length field */ + length = le64_to_cpu(disk_strm->length); - /* Length including neither the null terminator nor the padding - * */ - length_no_padding = WIM_ADS_ENTRY_DISK_SIZE + - cur_entry->stream_name_len; + /* 8-byte align the length */ + length = (length + 7) & ~7; - /* Length including the null terminator and the padding */ - total_length = ((length_no_padding + 2) + 7) & ~7; + /* Make sure the length field is neither so small it doesn't + * include all the fixed-length data nor so large it overflows + * the metadata resource buffer. */ + if (length < sizeof(struct wim_extra_stream_entry_on_disk) || + length > (end - p)) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - wimlib_assert(total_length == ads_entry_total_length(cur_entry)); + /* Read the rest of the fixed-length data. */ - if (remaining_size < length_no_padding) { - ERROR("Stream entries go past end of metadata resource"); - ERROR("(remaining_size = %"PRIu64" bytes, " - "length_no_padding = %"PRIu64" bytes)", - remaining_size, length_no_padding); - ret = WIMLIB_ERR_INVALID_DENTRY; - goto out_free_ads_entries; - } + copy_hash(strm->_stream_hash, disk_strm->hash); + name_nbytes = le16_to_cpu(disk_strm->name_nbytes); - /* The @length field in the on-disk ADS entry is expected to be - * equal to @total_length, which includes all of the entry and - * the padding that follows it to align the next ADS entry to an - * 8-byte boundary. However, to be safe, we'll accept the - * length field as long as it's not less than the un-padded - * total length and not more than the padded total length. */ - if (length < length_no_padding || length > total_length) { - ERROR("Stream entry has unexpected length " - "field (length field = %"PRIu64", " - "unpadded total length = %"PRIu64", " - "padded total length = %"PRIu64")", - length, length_no_padding, total_length); - ret = WIMLIB_ERR_INVALID_DENTRY; - goto out_free_ads_entries; - } + /* If stream_name_nbytes != 0, the stream is named. */ + if (name_nbytes != 0) { + /* The name is encoded in UTF16-LE, which uses 2-byte + * coding units, so the length of the name had better be + * an even number of bytes. */ + if (name_nbytes & 1) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - if (cur_entry->stream_name_len) { - cur_entry->stream_name = MALLOC(cur_entry->stream_name_len); - if (!cur_entry->stream_name) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ads_entries; - } - get_bytes(p, cur_entry->stream_name_len, - (u8*)cur_entry->stream_name); - cur_entry->stream_name_utf8 = utf16_to_utf8(cur_entry->stream_name, - cur_entry->stream_name_len, - &utf8_len); - cur_entry->stream_name_utf8_len = utf8_len; - - if (!cur_entry->stream_name_utf8) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ads_entries; - } + /* Add the length of the stream name to get the length + * we actually need to read. Make sure this isn't more + * than the specified length of the entry. */ + if (sizeof(struct wim_extra_stream_entry_on_disk) + + name_nbytes > length) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; + + strm->stream_name = utf16le_dupz(disk_strm->name, + name_nbytes); + if (!strm->stream_name) + return WIMLIB_ERR_NOMEM; + } else { + strm->stream_name = (utf16lechar *)NO_STREAM_NAME; } - /* It's expected that the size of every ADS entry is a multiple - * of 8. However, to be safe, I'm allowing the possibility of - * an ADS entry at the very end of the metadata resource ending - * un-aligned. So although we still need to increment the input - * pointer by @total_length to reach the next ADS entry, it's - * possible that less than @total_length is actually remaining - * in the metadata resource. We should set the remaining size to - * 0 bytes if this happens. */ - p = p_save + total_length; - if (remaining_size < total_length) - remaining_size = 0; - else - remaining_size -= total_length; + + strm->stream_type = STREAM_TYPE_UNKNOWN; + + p += length; } - inode->ads_entries = ads_entries; + + inode->i_next_stream_id = inode->i_num_streams; + + /* Now, assign a type to each stream. Unfortunately this requires + * various hacks because stream types aren't explicitly provided in the + * WIM on-disk format. */ + + if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) + assign_stream_types_encrypted(inode); + else + assign_stream_types_unencrypted(inode); + + *offset_p += p - orig_p; return 0; -out_free_ads_entries: - for (u16 i = 0; i < num_ads; i++) - free_ads_entry(ads_entries[i]); - FREE(ads_entries); - return ret; } -/* - * Reads a directory entry, including all alternate data stream entries that - * follow it, from the WIM image's metadata resource. - * - * @metadata_resource: Buffer containing the uncompressed metadata resource. - * @metadata_resource_len: Length of the metadata resource. - * @offset: Offset of this directory entry in the metadata resource. - * @dentry: A `struct dentry' that will be filled in by this function. - * - * Return 0 on success or nonzero on failure. On failure, @dentry have been - * modified, bu it will be left with no pointers to any allocated buffers. - * On success, the dentry->length field must be examined. If zero, this was a - * special "end of directory" dentry and not a real dentry. If nonzero, this - * was a real dentry. - */ -int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, - u64 offset, struct dentry *dentry) +/* Read a dentry, including all extra stream entries that follow it, from an + * uncompressed metadata resource buffer. */ +static int +read_dentry(const u8 * restrict buf, size_t buf_len, + u64 *offset_p, struct wim_dentry **dentry_ret) { + u64 offset = *offset_p; + u64 length; const u8 *p; + const struct wim_dentry_on_disk *disk_dentry; + struct wim_dentry *dentry; + struct wim_inode *inode; + u16 short_name_nbytes; + u16 file_name_nbytes; u64 calculated_size; - char *file_name = NULL; - char *file_name_utf8 = NULL; - char *short_name = NULL; - u16 short_name_len; - u16 file_name_len; - size_t file_name_utf8_len = 0; int ret; - struct inode *inode = NULL; - - dentry_common_init(dentry); - /*Make sure the dentry really fits into the metadata resource.*/ - if (offset + 8 > metadata_resource_len || offset + 8 < offset) { - ERROR("Directory entry starting at %"PRIu64" ends past the " - "end of the metadata resource (size %"PRIu64")", - offset, metadata_resource_len); - return WIMLIB_ERR_INVALID_DENTRY; - } + BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE); /* Before reading the whole dentry, we need to read just the length. * This is because a dentry of length 8 (that is, just the length field) * terminates the list of sibling directory entries. */ - p = get_u64(&metadata_resource[offset], &dentry->length); + /* Check for buffer overrun. */ + if (unlikely(offset + sizeof(u64) > buf_len || + offset + sizeof(u64) < offset)) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - /* A zero length field (really a length of 8, since that's how big the - * directory entry is...) indicates that this is the end of directory - * dentry. We do not read it into memory as an actual dentry, so just - * return successfully in that case. */ - if (dentry->length == 0) - return 0; + /* Get pointer to the dentry data. */ + p = &buf[offset]; + disk_dentry = (const struct wim_dentry_on_disk*)p; - /* If the dentry does not overflow the metadata resource buffer and is - * not too short, read the rest of it (excluding the alternate data - * streams, but including the file name and short name variable-length - * fields) into memory. */ - if (offset + dentry->length >= metadata_resource_len - || offset + dentry->length < offset) - { - ERROR("Directory entry at offset %"PRIu64" and with size " - "%"PRIu64" ends past the end of the metadata resource " - "(size %"PRIu64")", - offset, dentry->length, metadata_resource_len); - return WIMLIB_ERR_INVALID_DENTRY; - } + /* Get dentry length. */ + length = (le64_to_cpu(disk_dentry->length) + 7) & ~7; - if (dentry->length < WIM_DENTRY_DISK_SIZE) { - ERROR("Directory entry has invalid length of %"PRIu64" bytes", - dentry->length); - return WIMLIB_ERR_INVALID_DENTRY; + /* Check for end-of-directory. */ + if (length <= 8) { + *dentry_ret = NULL; + return 0; } - inode = new_timeless_inode(); - if (!inode) - return WIMLIB_ERR_NOMEM; + /* Validate dentry length. */ + if (unlikely(length < sizeof(struct wim_dentry_on_disk))) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - p = get_u32(p, &inode->attributes); - p = get_u32(p, (u32*)&inode->security_id); - p = get_u64(p, &dentry->subdir_offset); + /* Check for buffer overrun. */ + if (unlikely(offset + length > buf_len || + offset + length < offset)) + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - /* 2 unused fields */ - p += 2 * sizeof(u64); - /*p = get_u64(p, &dentry->unused1);*/ - /*p = get_u64(p, &dentry->unused2);*/ + /* Allocate new dentry structure, along with a preliminary inode. */ + ret = new_dentry_with_new_inode(NULL, false, &dentry); + if (ret) + return ret; - p = get_u64(p, &inode->creation_time); - p = get_u64(p, &inode->last_access_time); - p = get_u64(p, &inode->last_write_time); + inode = dentry->d_inode; - p = get_bytes(p, SHA1_HASH_SIZE, inode->hash); - - /* - * I don't know what's going on here. It seems like M$ screwed up the + /* Read more fields: some into the dentry, and some into the inode. */ + inode->i_attributes = le32_to_cpu(disk_dentry->attributes); + inode->i_security_id = le32_to_cpu(disk_dentry->security_id); + dentry->subdir_offset = le64_to_cpu(disk_dentry->subdir_offset); + inode->i_creation_time = le64_to_cpu(disk_dentry->creation_time); + inode->i_last_access_time = le64_to_cpu(disk_dentry->last_access_time); + inode->i_last_write_time = le64_to_cpu(disk_dentry->last_write_time); + + /* I don't know what's going on here. It seems like M$ screwed up the * reparse points, then put the fields in the same place and didn't - * document it. The WIM_HDR_FLAG_RP_FIX flag in the WIM header might - * have something to do with this, but it's not documented. - */ - if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) { - /* ??? */ - p += 4; - p = get_u32(p, &inode->reparse_tag); - p += 4; + * document it. So we have some fields we read for reparse points, and + * some fields in the same place for non-reparse-points. */ + if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) { + inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->reparse.rp_unknown_1); + inode->i_reparse_tag = le32_to_cpu(disk_dentry->reparse.reparse_tag); + inode->i_rp_unknown_2 = le16_to_cpu(disk_dentry->reparse.rp_unknown_2); + inode->i_not_rpfixed = le16_to_cpu(disk_dentry->reparse.not_rpfixed); + /* Leave inode->i_ino at 0. Note: this means that WIM cannot + * represent multiple hard links to a reparse point file. */ } else { - p = get_u32(p, &inode->reparse_tag); - p = get_u64(p, &inode->ino); + inode->i_rp_unknown_1 = le32_to_cpu(disk_dentry->nonreparse.rp_unknown_1); + inode->i_ino = le64_to_cpu(disk_dentry->nonreparse.hard_link_group_id); } - /* By the way, the reparse_reserved field does not actually exist (at - * least when the file is not a reparse point) */ - - p = get_u16(p, &inode->num_ads); + /* Now onto reading the names. There are two of them: the (long) file + * name, and the short name. */ - p = get_u16(p, &short_name_len); - p = get_u16(p, &file_name_len); + short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes); + file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes); - /* We now know the length of the file name and short name. Make sure - * the length of the dentry is large enough to actually hold them. - * - * The calculated length here is unaligned to allow for the possibility - * that the dentry->length names an unaligned length, although this - * would be unexpected. */ - calculated_size = __dentry_correct_length_unaligned(file_name_len, - short_name_len); - - if (dentry->length < calculated_size) { - ERROR("Unexpected end of directory entry! (Expected " - "at least %"PRIu64" bytes, got %"PRIu64" bytes. " - "short_name_len = %hu, file_name_len = %hu)", - calculated_size, dentry->length, - short_name_len, file_name_len); - return WIMLIB_ERR_INVALID_DENTRY; + if (unlikely((short_name_nbytes & 1) | (file_name_nbytes & 1))) { + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; + goto err_free_dentry; } - /* Read the filename if present. Note: if the filename is empty, there - * is no null terminator following it. */ - if (file_name_len) { - file_name = MALLOC(file_name_len); - if (!file_name) { - ERROR("Failed to allocate %hu bytes for dentry file name", - file_name_len); - return WIMLIB_ERR_NOMEM; - } - p = get_bytes(p, file_name_len, file_name); - - /* Convert filename to UTF-8. */ - file_name_utf8 = utf16_to_utf8(file_name, file_name_len, - &file_name_utf8_len); + /* We now know the length of the file name and short name. Make sure + * the length of the dentry is large enough to actually hold them. */ + calculated_size = dentry_min_len_with_names(file_name_nbytes, + short_name_nbytes); - if (!file_name_utf8) { - ERROR("Failed to allocate memory to convert UTF-16 " - "filename (%hu bytes) to UTF-8", file_name_len); - ret = WIMLIB_ERR_NOMEM; - goto out_free_file_name; - } - if (*(u16*)p) - WARNING("Expected two zero bytes following the file name " - "`%s', but found non-zero bytes", file_name_utf8); - p += 2; + if (unlikely(length < calculated_size)) { + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; + goto err_free_dentry; } - /* Align the calculated size */ - calculated_size = (calculated_size + 7) & ~7; + /* Advance p to point past the base dentry, to the first name. */ + p += sizeof(struct wim_dentry_on_disk); - if (dentry->length > calculated_size) { - /* Weird; the dentry says it's longer than it should be. Note - * that the length field does NOT include the size of the - * alternate stream entries. */ - - /* Strangely, some directory entries inexplicably have a little - * over 70 bytes of extra data. The exact amount of data seems - * to be 72 bytes, but it is aligned on the next 8-byte - * boundary. It does NOT seem to be alternate data stream - * entries. Here's an example of the aligned data: - * - * 01000000 40000000 6c786bba c58ede11 b0bb0026 1870892a b6adb76f - * e63a3e46 8fca8653 0d2effa1 6c786bba c58ede11 b0bb0026 1870892a - * 00000000 00000000 00000000 00000000 - * - * Here's one interpretation of how the data is laid out. - * - * struct unknown { - * u32 field1; (always 0x00000001) - * u32 field2; (always 0x40000000) - * u8 data[48]; (???) - * u64 reserved1; (always 0) - * u64 reserved2; (always 0) - * };*/ - DEBUG("Dentry for file or directory `%s' has %zu extra " - "bytes of data", - file_name_utf8, dentry->length - calculated_size); + /* Read the filename if present. Note: if the filename is empty, there + * is no null terminator following it. */ + if (file_name_nbytes) { + dentry->file_name = utf16le_dupz(p, file_name_nbytes); + if (dentry->file_name == NULL) { + ret = WIMLIB_ERR_NOMEM; + goto err_free_dentry; + } + dentry->file_name_nbytes = file_name_nbytes; + p += (u32)file_name_nbytes + 2; } /* Read the short filename if present. Note: if there is no short * filename, there is no null terminator following it. */ - if (short_name_len) { - short_name = MALLOC(short_name_len); - if (!short_name) { - ERROR("Failed to allocate %hu bytes for short filename", - short_name_len); + if (short_name_nbytes) { + dentry->short_name = utf16le_dupz(p, short_name_nbytes); + if (dentry->short_name == NULL) { ret = WIMLIB_ERR_NOMEM; - goto out_free_file_name_utf8; - } - - p = get_bytes(p, short_name_len, short_name); - if (*(u16*)p) - WARNING("Expected two zero bytes following the file name " - "`%s', but found non-zero bytes", file_name_utf8); - p += 2; - } - - /* - * Read the alternate data streams, if present. dentry->num_ads tells - * us how many they are, and they will directly follow the dentry - * on-disk. - * - * Note that each alternate data stream entry begins on an 8-byte - * aligned boundary, and the alternate data stream entries are NOT - * included in the dentry->length field for some reason. - */ - if (inode->num_ads != 0) { - if (calculated_size > metadata_resource_len - offset) { - ERROR("Not enough space in metadata resource for " - "alternate stream entries"); - ret = WIMLIB_ERR_INVALID_DENTRY; - goto out_free_short_name; + goto err_free_dentry; } - ret = read_ads_entries(&metadata_resource[offset + calculated_size], - inode, - metadata_resource_len - offset - calculated_size); - if (ret != 0) - goto out_free_short_name; + dentry->short_name_nbytes = short_name_nbytes; + p += (u32)short_name_nbytes + 2; } - /* We've read all the data for this dentry. Set the names and their - * lengths, and we've done. */ - dentry->inode = inode; - dentry->file_name = file_name; - dentry->file_name_utf8 = file_name_utf8; - dentry->short_name = short_name; - dentry->file_name_len = file_name_len; - dentry->file_name_utf8_len = file_name_utf8_len; - dentry->short_name_len = short_name_len; + /* Read extra data at end of dentry (but before extra stream entries). + * This may contain tagged metadata items. */ + ret = read_extra_data(p, &buf[offset + length], inode); + if (ret) + goto err_free_dentry; + + offset += length; + + /* Set up the inode's collection of streams. */ + ret = setup_inode_streams(&buf[offset], + &buf[buf_len], + inode, + le16_to_cpu(disk_dentry->num_extra_streams), + disk_dentry->default_hash, + &offset); + if (ret) + goto err_free_dentry; + + *offset_p = offset; /* Sets offset of next dentry in directory */ + *dentry_ret = dentry; return 0; -out_free_short_name: - FREE(short_name); -out_free_file_name_utf8: - FREE(file_name_utf8); -out_free_file_name: - FREE(file_name); -out_free_inode: - free_inode(inode); + +err_free_dentry: + free_dentry(dentry); return ret; } -/* Run some miscellaneous verifications on a WIM dentry */ -int verify_dentry(struct dentry *dentry, void *wim) +/* Is the dentry named "." or ".." ? */ +static bool +dentry_is_dot_or_dotdot(const struct wim_dentry *dentry) { - const WIMStruct *w = wim; - const struct lookup_table *table = w->lookup_table; - const struct wim_security_data *sd = wim_const_security_data(w); - const struct inode *inode = dentry->inode; - int ret = WIMLIB_ERR_INVALID_DENTRY; - - /* Check the security ID */ - if (inode->security_id < -1) { - ERROR("Dentry `%s' has an invalid security ID (%d)", - dentry->full_path_utf8, inode->security_id); - goto out; + if (dentry->file_name_nbytes <= 4) { + if (dentry->file_name_nbytes == 4) { + if (dentry->file_name[0] == cpu_to_le16('.') && + dentry->file_name[1] == cpu_to_le16('.')) + return true; + } else if (dentry->file_name_nbytes == 2) { + if (dentry->file_name[0] == cpu_to_le16('.')) + return true; + } } - if (inode->security_id >= sd->num_entries) { - ERROR("Dentry `%s' has an invalid security ID (%d) " - "(there are only %u entries in the security table)", - dentry->full_path_utf8, inode->security_id, - sd->num_entries); - goto out; + return false; +} + +static int +read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len, + struct wim_dentry * restrict dir) +{ + u64 cur_offset = dir->subdir_offset; + + /* Check for cyclic directory structure, which would cause infinite + * recursion if not handled. */ + for (struct wim_dentry *d = dir->d_parent; + !dentry_is_root(d); d = d->d_parent) + { + if (unlikely(d->subdir_offset == cur_offset)) { + ERROR("Cyclic directory structure detected: children " + "of \"%"TS"\" coincide with children of \"%"TS"\"", + dentry_full_path(dir), dentry_full_path(d)); + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; + } } - /* Check that lookup table entries for all the resources exist, except - * if the SHA1 message digest is all 0's, which indicates there is - * intentionally no resource there. */ - if (w->hdr.total_parts == 1) { - for (unsigned i = 0; i <= inode->num_ads; i++) { - struct lookup_table_entry *lte; - const u8 *hash; - hash = inode_stream_hash_unresolved(inode, i); - lte = __lookup_resource(table, hash); - if (!lte && !is_zero_hash(hash)) { - ERROR("Could not find lookup table entry for stream " - "%u of dentry `%s'", i, dentry->full_path_utf8); - goto out; + for (;;) { + struct wim_dentry *child; + struct wim_dentry *duplicate; + int ret; + + /* Read next child of @dir. */ + ret = read_dentry(buf, buf_len, &cur_offset, &child); + if (ret) + return ret; + + /* Check for end of directory. */ + if (child == NULL) + return 0; + + /* All dentries except the root should be named. */ + if (unlikely(!dentry_has_long_name(child))) { + WARNING("Ignoring unnamed dentry in " + "directory \"%"TS"\"", dentry_full_path(dir)); + free_dentry(child); + continue; + } + + /* Don't allow files named "." or "..". */ + if (unlikely(dentry_is_dot_or_dotdot(child))) { + WARNING("Ignoring file named \".\" or \"..\"; " + "potentially malicious archive!!!"); + free_dentry(child); + continue; + } + + /* Link the child into the directory. */ + duplicate = dentry_add_child(dir, child); + if (unlikely(duplicate)) { + /* We already found a dentry with this same + * case-sensitive long name. Only keep the first one. + */ + WARNING("Ignoring duplicate file \"%"TS"\" " + "(the WIM image already contains a file " + "at that path with the exact same name)", + dentry_full_path(duplicate)); + free_dentry(child); + continue; + } + + /* If this child is a directory that itself has children, call + * this procedure recursively. */ + if (child->subdir_offset != 0) { + if (likely(dentry_is_directory(child))) { + ret = read_dentry_tree_recursive(buf, + buf_len, + child); + if (ret) + return ret; + } else { + WARNING("Ignoring children of " + "non-directory file \"%"TS"\"", + dentry_full_path(child)); } } } +} - /* Make sure there is only one un-named stream. */ - unsigned num_unnamed_streams = 0; - for (unsigned i = 0; i <= inode->num_ads; i++) { - const u8 *hash; - hash = inode_stream_hash_unresolved(inode, i); - if (!inode_stream_name_len(inode, i) && !is_zero_hash(hash)) - num_unnamed_streams++; - } - if (num_unnamed_streams > 1) { - ERROR("Dentry `%s' has multiple (%u) un-named streams", - dentry->full_path_utf8, num_unnamed_streams); - goto out; - } +/* + * Read a tree of dentries from a WIM metadata resource. + * + * @buf: + * Buffer containing an uncompressed WIM metadata resource. + * + * @buf_len: + * Length of the uncompressed metadata resource, in bytes. + * + * @root_offset + * Offset in the metadata resource of the root of the dentry tree. + * + * @root_ret: + * On success, either NULL or a pointer to the root dentry is written to + * this location. The former case only occurs in the unexpected case that + * the tree began with an end-of-directory entry. + * + * Return values: + * WIMLIB_ERR_SUCCESS (0) + * WIMLIB_ERR_INVALID_METADATA_RESOURCE + * WIMLIB_ERR_NOMEM + */ +int +read_dentry_tree(const u8 *buf, size_t buf_len, + u64 root_offset, struct wim_dentry **root_ret) +{ + int ret; + struct wim_dentry *root; - /* Cannot have a short name but no long name */ - if (dentry->short_name_len && !dentry->file_name_len) { - ERROR("Dentry `%s' has a short name but no long name", - dentry->full_path_utf8); - goto out; - } + DEBUG("Reading dentry tree (root_offset=%"PRIu64")", root_offset); + + ret = read_dentry(buf, buf_len, &root_offset, &root); + if (ret) + return ret; - /* Make sure root dentry is unnamed */ - if (dentry_is_root(dentry)) { - if (dentry->file_name_len) { - ERROR("The root dentry is named `%s', but it must " - "be unnamed", dentry->file_name_utf8); - goto out; + if (likely(root != NULL)) { + if (unlikely(dentry_has_long_name(root) || + dentry_has_short_name(root))) + { + WARNING("The root directory has a nonempty name; " + "removing it."); + dentry_set_name(root, NULL); } - } -#if 0 - /* Check timestamps */ - if (inode->last_access_time < inode->creation_time || - inode->last_write_time < inode->creation_time) { - WARNING("Dentry `%s' was created after it was last accessed or " - "written to", dentry->full_path_utf8); + if (unlikely(!dentry_is_directory(root))) { + ERROR("The root of the WIM image is not a directory!"); + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; + goto err_free_dentry_tree; + } + + if (likely(root->subdir_offset != 0)) { + ret = read_dentry_tree_recursive(buf, buf_len, root); + if (ret) + goto err_free_dentry_tree; + } + } else { + WARNING("The metadata resource has no directory entries; " + "treating as an empty image."); } -#endif + *root_ret = root; + return 0; - ret = 0; -out: +err_free_dentry_tree: + free_dentry_tree(root, NULL); return ret; } -/* - * Writes a WIM dentry to an output buffer. - * - * @dentry: The dentry structure. - * @p: The memory location to write the data to. - * @return: Pointer to the byte after the last byte we wrote as part of the - * dentry. - */ -static u8 *write_dentry(const struct dentry *dentry, u8 *p) +static u8 * +write_extra_stream_entry(u8 * restrict p, const utf16lechar * restrict name, + const u8 * restrict hash) { + struct wim_extra_stream_entry_on_disk *disk_strm = + (struct wim_extra_stream_entry_on_disk *)p; u8 *orig_p = p; - const u8 *hash; - const struct inode *inode = dentry->inode; - - /* We calculate the correct length of the dentry ourselves because the - * dentry->length field may been set to an unexpected value from when we - * read the dentry in (for example, there may have been unknown data - * appended to the end of the dentry...) */ - u64 length = dentry_correct_length(dentry); - - p = put_u64(p, length); - p = put_u32(p, inode->attributes); - p = put_u32(p, inode->security_id); - p = put_u64(p, dentry->subdir_offset); - p = put_u64(p, 0); /* unused1 */ - p = put_u64(p, 0); /* unused2 */ - p = put_u64(p, inode->creation_time); - p = put_u64(p, inode->last_access_time); - p = put_u64(p, inode->last_write_time); - hash = inode_stream_hash(inode, 0); - p = put_bytes(p, SHA1_HASH_SIZE, hash); - if (inode->attributes & FILE_ATTRIBUTE_REPARSE_POINT) { - p = put_zeroes(p, 4); - p = put_u32(p, inode->reparse_tag); - p = put_zeroes(p, 4); - } else { - u64 link_group_id; - p = put_u32(p, 0); - if (inode->link_count == 1) - link_group_id = 0; - else - link_group_id = inode->ino; - p = put_u64(p, link_group_id); - } - p = put_u16(p, inode->num_ads); - p = put_u16(p, dentry->short_name_len); - p = put_u16(p, dentry->file_name_len); - if (dentry->file_name_len) { - p = put_bytes(p, dentry->file_name_len, (u8*)dentry->file_name); - p = put_u16(p, 0); /* filename padding, 2 bytes. */ - } - if (dentry->short_name) { - p = put_bytes(p, dentry->short_name_len, (u8*)dentry->short_name); - p = put_u16(p, 0); /* short name padding, 2 bytes */ - } + size_t name_nbytes; + if (name == NO_STREAM_NAME) + name_nbytes = 0; + else + name_nbytes = utf16le_len_bytes(name); + + disk_strm->reserved = 0; + copy_hash(disk_strm->hash, hash); + disk_strm->name_nbytes = cpu_to_le16(name_nbytes); + p += sizeof(struct wim_extra_stream_entry_on_disk); + if (name_nbytes != 0) + p = mempcpy(p, name, name_nbytes + 2); /* Align to 8-byte boundary */ - wimlib_assert(length >= (p - orig_p) - && length - (p - orig_p) <= 7); - p = put_zeroes(p, length - (p - orig_p)); - - /* Write the alternate data streams, if there are any. Please see - * read_ads_entries() for comments about the format of the on-disk - * alternate data stream entries. */ - for (u16 i = 0; i < inode->num_ads; i++) { - p = put_u64(p, ads_entry_total_length(inode->ads_entries[i])); - p = put_u64(p, 0); /* Unused */ - hash = inode_stream_hash(inode, i + 1); - p = put_bytes(p, SHA1_HASH_SIZE, hash); - p = put_u16(p, inode->ads_entries[i]->stream_name_len); - if (inode->ads_entries[i]->stream_name_len) { - p = put_bytes(p, inode->ads_entries[i]->stream_name_len, - (u8*)inode->ads_entries[i]->stream_name); - p = put_u16(p, 0); - } - p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8); - } -#ifdef ENABLE_ASSERTIONS - wimlib_assert(p - orig_p == __dentry_total_length(dentry, length)); -#endif + while ((uintptr_t)p & 7) + *p++ = 0; + disk_strm->length = cpu_to_le64(p - orig_p); return p; } -/* Recursive function that writes a dentry tree rooted at @parent, not including - * @parent itself, which has already been written. */ -static u8 *write_dentry_tree_recursive(const struct dentry *parent, u8 *p) +/* + * Write a WIM dentry to an output buffer. + * + * This includes any extra stream entries that may follow the dentry itself. + * + * @dentry: + * The dentry to write. + * + * @p: + * The memory location to which to write the data. + * + * Returns a pointer to the byte following the last written. + */ +static u8 * +write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p) { - const struct dentry *child; + const struct wim_inode *inode; + struct wim_dentry_on_disk *disk_dentry; + const u8 *orig_p; + + wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */ + orig_p = p; + + inode = dentry->d_inode; + disk_dentry = (struct wim_dentry_on_disk*)p; + + disk_dentry->attributes = cpu_to_le32(inode->i_attributes); + disk_dentry->security_id = cpu_to_le32(inode->i_security_id); + disk_dentry->subdir_offset = cpu_to_le64(dentry->subdir_offset); + + disk_dentry->unused_1 = cpu_to_le64(0); + disk_dentry->unused_2 = cpu_to_le64(0); + + disk_dentry->creation_time = cpu_to_le64(inode->i_creation_time); + disk_dentry->last_access_time = cpu_to_le64(inode->i_last_access_time); + disk_dentry->last_write_time = cpu_to_le64(inode->i_last_write_time); + if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) { + disk_dentry->reparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1); + disk_dentry->reparse.reparse_tag = cpu_to_le32(inode->i_reparse_tag); + disk_dentry->reparse.rp_unknown_2 = cpu_to_le16(inode->i_rp_unknown_2); + disk_dentry->reparse.not_rpfixed = cpu_to_le16(inode->i_not_rpfixed); + } else { + disk_dentry->nonreparse.rp_unknown_1 = cpu_to_le32(inode->i_rp_unknown_1); + disk_dentry->nonreparse.hard_link_group_id = + cpu_to_le64((inode->i_nlink == 1) ? 0 : inode->i_ino); + } - /* Nothing to do if this dentry has no children. */ - if (parent->subdir_offset == 0) - return p; + disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes); + disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes); + p += sizeof(struct wim_dentry_on_disk); - /* Write child dentries and end-of-directory entry. - * - * Note: we need to write all of this dentry's children before - * recursively writing the directory trees rooted at each of the child - * dentries, since the on-disk dentries for a dentry's children are - * always located at consecutive positions in the metadata resource! */ - child = parent->inode->children; - if (child) { - do { - p = write_dentry(child, p); - child = child->next; - } while (child != parent->inode->children); + wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry)); + + if (dentry_has_long_name(dentry)) + p = mempcpy(p, dentry->file_name, (u32)dentry->file_name_nbytes + 2); + + if (dentry_has_short_name(dentry)) + p = mempcpy(p, dentry->short_name, (u32)dentry->short_name_nbytes + 2); + + /* Align to 8-byte boundary */ + while ((uintptr_t)p & 7) + *p++ = 0; + + if (inode->i_extra_size) { + /* Extra tagged items --- not usually present. */ + p = mempcpy(p, inode->i_extra, inode->i_extra_size); + + /* Align to 8-byte boundary */ + while ((uintptr_t)p & 7) + *p++ = 0; } - /* write end of directory entry */ - p = put_u64(p, 0); + disk_dentry->length = cpu_to_le64(p - orig_p); - /* Recurse on children. */ - if (child) { - do { - p = write_dentry_tree_recursive(child, p); - child = child->next; - } while (child != parent->inode->children); + /* Streams */ + + if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) { + const struct wim_inode_stream *efs_strm; + const u8 *efs_hash; + + efs_strm = inode_get_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA, + NO_STREAM_NAME); + efs_hash = efs_strm ? stream_hash(efs_strm) : zero_hash; + copy_hash(disk_dentry->default_hash, efs_hash); + disk_dentry->num_extra_streams = cpu_to_le16(0); + } else { + /* + * Extra stream entries: + * + * - Use one extra stream entry for each named data stream + * - Use one extra stream entry for the unnamed data stream when there is either: + * - a reparse point stream + * - at least one named data stream (for Windows PE bug workaround) + * - Use one extra stream entry for the reparse point stream if there is one + */ + bool have_named_data_stream = false; + bool have_reparse_point_stream = false; + u16 num_extra_streams = 0; + const u8 *unnamed_data_stream_hash = zero_hash; + const u8 *reparse_point_hash; + for (unsigned i = 0; i < inode->i_num_streams; i++) { + const struct wim_inode_stream *strm = &inode->i_streams[i]; + if (strm->stream_type == STREAM_TYPE_DATA) { + if (stream_is_named(strm)) + have_named_data_stream = true; + else + unnamed_data_stream_hash = stream_hash(strm); + } else if (strm->stream_type == STREAM_TYPE_REPARSE_POINT) { + have_reparse_point_stream = true; + reparse_point_hash = stream_hash(strm); + } + } + + if (have_reparse_point_stream || have_named_data_stream) { + + copy_hash(disk_dentry->default_hash, zero_hash); + + if (have_reparse_point_stream) { + p = write_extra_stream_entry(p, NO_STREAM_NAME, + reparse_point_hash); + num_extra_streams++; + } + + p = write_extra_stream_entry(p, NO_STREAM_NAME, + unnamed_data_stream_hash); + num_extra_streams++; + } else { + copy_hash(disk_dentry->default_hash, unnamed_data_stream_hash); + } + + for (unsigned i = 0; i < inode->i_num_streams; i++) { + const struct wim_inode_stream *strm = &inode->i_streams[i]; + if (stream_is_named_data_stream(strm)) { + p = write_extra_stream_entry(p, strm->stream_name, + stream_hash(strm)); + num_extra_streams++; + } + } + disk_dentry->num_extra_streams = cpu_to_le16(num_extra_streams); } + return p; } -/* Writes a directory tree to the metadata resource. - * - * @root: Root of the dentry tree. - * @p: Pointer to a buffer with enough space for the dentry tree. - * - * Returns pointer to the byte after the last byte we wrote. - */ -u8 *write_dentry_tree(const struct dentry *root, u8 *p) +static int +write_dir_dentries(struct wim_dentry *dir, void *_pp) { - wimlib_assert(dentry_is_root(root)); - - /* If we're the root dentry, we have no parent that already - * wrote us, so we need to write ourselves. */ - p = write_dentry(root, p); + if (dir->subdir_offset != 0) { + u8 **pp = _pp; + u8 *p = *pp; + struct wim_dentry *child; - /* Write end of directory entry after the root dentry just to be safe; - * however the root dentry obviously cannot have any siblings. */ - p = put_u64(p, 0); + /* write child dentries */ + for_dentry_child(child, dir) + p = write_dentry(child, p); - /* Recursively write the rest of the dentry tree. */ - return write_dentry_tree_recursive(root, p); + /* write end of directory entry */ + *(u64*)p = 0; + p += 8; + *pp = p; + } + return 0; } -/* Reads the children of a dentry, and all their children, ..., etc. from the - * metadata resource and into the dentry tree. - * - * @metadata_resource: An array that contains the uncompressed metadata - * resource for the WIM file. +/* + * Write a directory tree to the metadata resource. * - * @metadata_resource_len: The length of the uncompressed metadata resource, in - * bytes. + * @root: + * The root of a dentry tree on which calculate_subdir_offsets() has been + * called. This cannot be NULL; if the dentry tree is empty, the caller is + * expected to first generate a dummy root directory. * - * @dentry: A pointer to a `struct dentry' that is the root of the directory - * tree and has already been read from the metadata resource. It - * does not need to be the real root because this procedure is - * called recursively. + * @p: + * Pointer to a buffer with enough space for the dentry tree. This size + * must have been obtained by calculate_subdir_offsets(). * - * @return: Zero on success, nonzero on failure. + * Returns a pointer to the byte following the last written. */ -int read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len, - struct dentry *dentry) +u8 * +write_dentry_tree(struct wim_dentry *root, u8 *p) { - u64 cur_offset = dentry->subdir_offset; - struct dentry *prev_child = NULL; - struct dentry *first_child = NULL; - struct dentry *child; - struct dentry cur_child; - int ret; + DEBUG("Writing dentry tree."); - /* - * If @dentry has no child dentries, nothing more needs to be done for - * this branch. This is the case for regular files, symbolic links, and - * *possibly* empty directories (although an empty directory may also - * have one child dentry that is the special end-of-directory dentry) - */ - if (cur_offset == 0) - return 0; - - /* Find and read all the children of @dentry. */ - while (1) { - - /* Read next child of @dentry into @cur_child. */ - ret = read_dentry(metadata_resource, metadata_resource_len, - cur_offset, &cur_child); - if (ret != 0) - break; - - /* Check for end of directory. */ - if (cur_child.length == 0) - break; - - /* Not end of directory. Allocate this child permanently and - * link it to the parent and previous child. */ - child = MALLOC(sizeof(struct dentry)); - if (!child) { - ERROR("Failed to allocate %zu bytes for new dentry", - sizeof(struct dentry)); - ret = WIMLIB_ERR_NOMEM; - break; - } - memcpy(child, &cur_child, sizeof(struct dentry)); - - if (prev_child) { - prev_child->next = child; - child->prev = prev_child; - } else { - first_child = child; - } - - child->parent = dentry; - prev_child = child; - list_add(&child->inode_dentry_list, &child->inode->dentry_list); + wimlib_assert(root != NULL); - /* If there are children of this child, call this procedure - * recursively. */ - if (child->subdir_offset != 0) { - ret = read_dentry_tree(metadata_resource, - metadata_resource_len, child); - if (ret != 0) - break; - } + /* write root dentry and end-of-directory entry following it */ + p = write_dentry(root, p); + *(u64*)p = 0; + p += 8; - /* Advance to the offset of the next child. Note: We need to - * advance by the TOTAL length of the dentry, not by the length - * child->length, which although it does take into account the - * padding, it DOES NOT take into account alternate stream - * entries. */ - cur_offset += dentry_total_length(child); - } + /* write the rest of the dentry tree */ + for_dentry_in_tree(root, write_dir_dentries, &p); - /* Link last child to first one, and set parent's children pointer to - * the first child. */ - if (prev_child) { - prev_child->next = first_child; - first_child->prev = prev_child; - } - dentry->inode->children = first_child; - return ret; + return p; }