X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fdentry.c;h=49c0cdf14784337e49b289b298eb4bff198ee251;hp=402c780f43f74295678dccd92f945e1283666e28;hb=ee547cc83f231d727e4d9984c23e86d96d3da769;hpb=c25d010093a6c897cca3fddc0cb1155a3786b664 diff --git a/src/dentry.c b/src/dentry.c index 402c780f..49c0cdf1 100644 --- a/src/dentry.c +++ b/src/dentry.c @@ -1,70 +1,93 @@ /* - * dentry.c + * dentry.c - see description below + */ + +/* + * Copyright (C) 2012, 2013, 2014, 2015 Eric Biggers + * + * 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. + * + * 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. * - * 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. + * 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/. */ /* - * Copyright (C) 2012, 2013, 2014 Eric Biggers + * This file contains logic to deal with WIM directory entries, or "dentries": * - * This file is part of wimlib, a library for working with WIM files. + * - 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 * - * 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. + * Notes: * - * 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. + * - 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. * - * You should have received a copy of the GNU General Public License along with - * wimlib; if not, see http://www.gnu.org/licenses/. + * - 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. + * + * - 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 "wimlib.h" -#include "wimlib/case.h" +#include + +#include "wimlib/assert.h" #include "wimlib/dentry.h" +#include "wimlib/inode.h" #include "wimlib/encoding.h" #include "wimlib/endianness.h" -#include "wimlib/error.h" -#include "wimlib/lookup_table.h" #include "wimlib/metadata.h" #include "wimlib/paths.h" -#include "wimlib/resource.h" -#include "wimlib/security.h" -#include "wimlib/sha1.h" -#include "wimlib/timestamp.h" - -#include /* 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 alternate - * data stream entries. Should be a multiple of 8 so that the following - * dentry or alternate data 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. + /* 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 situation, which is - * undocumented, indicates the end of a list of sibling nodes 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. */ + * 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; - /* Attributes of the file or directory. This is a bitwise OR of the - * FILE_ATTRIBUTE_* constants and should correspond to the value + /* 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; @@ -84,7 +107,6 @@ struct wim_dentry_on_disk { 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 @@ -93,250 +115,285 @@ struct wim_dentry_on_disk { le64 last_access_time; le64 last_write_time; - /* Vaguely, the SHA-1 message digest ("hash") of the file's contents. - * More specifically, this is for the "unnamed data stream" rather than - * any "alternate data streams". This hash value is used to look up the - * corresponding entry in the WIM's stream lookup table to actually find - * the file contents within the WIM. - * - * If the file has no unnamed data stream (e.g. is a directory), then - * this field will be all zeroes. If the unnamed data stream is empty - * (i.e. an "empty file"), then this field is also expected to be all - * zeroes. (It will be if wimlib created the WIM image, at least; - * otherwise it can't be ruled out that the SHA-1 message digest of 0 - * bytes of data is given explicitly.) + /* + * Usually this is the SHA-1 message digest of the file's "contents" + * (the unnamed data stream). * - * If the file has reparse data, then this field will instead specify - * the SHA-1 message digest of the reparse data. If it is somehow - * possible for a file to have both an unnamed data stream and reparse - * data, then this is not handled by wimlib. + * 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. * - * As a further special case, if this field is all zeroes but there is - * an alternate data stream entry with no name and a nonzero SHA-1 - * message digest field, then that hash must be used instead of this - * one. In fact, when named data streams are present, some versions of - * Windows PE contain a bug where they only look in the alternate data - * stream entries for the unnamed data stream, not here. + * 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 unnamed_stream_hash[SHA1_HASH_SIZE]; + u8 default_hash[SHA1_HASH_SIZE]; - /* The format of the following data is not yet completely known and they - * do not correspond to Microsoft's documentation. + /* Unknown field (maybe accidental padding) */ + le32 unknown_0x54; + + /* + * The following 8-byte union contains either information about the + * reparse point (for files with FILE_ATTRIBUTE_REPARSE_POINT set), or + * the "hard link group ID" (for other files). + * + * The reparse point information contains ReparseTag and ReparseReserved + * from the header of the reparse point buffer. It also contains a flag + * that indicates whether a reparse point fixup (for the target of an + * absolute symbolic link or junction) was done or not. * - * 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. + * The "hard link group ID" is like an inode number; all dentries for + * the same inode share the same value. See inode_fixup.c for more + * information. * - * 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. We have to - * handle this by using special code to use distinguishing features - * (which is possible because some information about the underlying - * inode is repeated in each dentry) to split up these fake hard link - * groups into what they actually are supposed to be. + * Note that this union creates the limitation that reparse point files + * cannot have multiple names (hard links). */ union { struct { - le32 rp_unknown_1; le32 reparse_tag; - le16 rp_unknown_2; - le16 not_rpfixed; + le16 rp_reserved; + le16 rp_flags; } _packed_attribute reparse; struct { - le32 rp_unknown_1; le64 hard_link_group_id; } _packed_attribute nonreparse; }; - /* Number of alternate data stream entries that directly follow this - * dentry on-disk. */ - le16 num_alternate_data_streams; + /* Number of extra stream entries that directly follow this dentry + * on-disk. */ + le16 num_extra_streams; - /* Length of this file's UTF-16LE encoded short name (8.3 DOS-compatible - * name), if present, in bytes, excluding the null terminator. If this - * file has no short name, then this field should be 0. */ + /* 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; - /* Length of this file's UTF-16LE encoded "long" name, excluding the - * null terminator. If this file has no short name, then this field - * should be 0. It's expected that only the root dentry has this field - * set to 0. */ - le16 file_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 name_nbytes; + + /* Beginning of optional, variable-length fields */ - /* Followed by variable length file name, in UTF16-LE, if - * file_name_nbytes != 0. Includes null terminator. */ - /*utf16lechar file_name[];*/ + /* If name_nbytes != 0, the next field will be the UTF-16LE encoded long + * name. This will be null-terminated, so the size of this field will + * really be name_nbytes + 2. */ + /*utf16lechar name[];*/ - /* Followed by variable length short name, in UTF16-LE, if - * short_name_nbytes != 0. Includes null terminator. */ + /* 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. + */ -/* Calculates the unaligned length, in bytes, of an on-disk WIM dentry that has - * a file name and short name that take the specified numbers of bytes. This - * excludes any alternate data stream entries that may follow the dentry. */ -static u64 -dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes) -{ - u64 length = sizeof(struct wim_dentry_on_disk); - if (file_name_nbytes) - length += file_name_nbytes + 2; - if (short_name_nbytes) - length += short_name_nbytes + 2; - return length; -} +/* 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 { -/* Calculates the unaligned length, in bytes, of an on-disk WIM dentry, based on - * the file name length and short name length. Note that dentry->length is - * ignored; also, this excludes any alternate data stream entries that may - * follow the dentry. */ -static u64 -dentry_correct_length_aligned(const struct wim_dentry *dentry) -{ - u64 len; + /* 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; - len = dentry_correct_length_unaligned(dentry->file_name_nbytes, - dentry->short_name_nbytes); - return (len + 7) & ~7; -} + /* Reserved field */ + le64 reserved; -static int -dentry_clear_short_name(struct wim_dentry *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]; + + /* 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 *name, + size_t name_nbytes) { + FREE(dentry->d_name); + dentry->d_name = name; + dentry->d_name_nbytes = name_nbytes; + if (dentry_has_short_name(dentry)) { - FREE(dentry->short_name); - dentry->short_name = NULL; - dentry->short_name_nbytes = 0; + FREE(dentry->d_short_name); + dentry->d_short_name = NULL; + dentry->d_short_name_nbytes = 0; } - return 0; } -/* Sets the name of a WIM dentry from a multibyte string. - * Only use this on dentries not inserted into the tree. Use rename_wim_path() - * to do a real rename. */ +/* + * 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(struct wim_dentry *dentry, const tchar *new_name) +dentry_set_name_utf16le(struct wim_dentry *dentry, const utf16lechar *name, + size_t name_nbytes) { - int ret; + utf16lechar *dup = NULL; - ret = get_utf16le_string(new_name, &dentry->file_name, - &dentry->file_name_nbytes); - if (ret) - return ret; - - return dentry_clear_short_name(dentry); + if (name_nbytes) { + dup = utf16le_dupz(name, name_nbytes); + if (!dup) + return WIMLIB_ERR_NOMEM; + } + do_dentry_set_name(dentry, dup, name_nbytes); + return 0; } -/* Sets the name of a WIM dentry from a UTF-16LE string. - * Only use this on dentries not inserted into the tree. Use rename_wim_path() - * to do a real rename. */ + +/* + * 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_utf16le(struct wim_dentry *dentry, const utf16lechar *new_name) +dentry_set_name(struct wim_dentry *dentry, const tchar *name) { - utf16lechar *name = NULL; - size_t name_nbytes = 0; - - if (new_name && *new_name) { - const utf16lechar *tmp; - - tmp = new_name; - do { - name_nbytes += sizeof(utf16lechar); - } while (*++tmp); + utf16lechar *name_utf16le = NULL; + size_t name_utf16le_nbytes = 0; + int ret; - name = memdup(new_name, name_nbytes + sizeof(utf16lechar)); - if (!name) - return WIMLIB_ERR_NOMEM; + if (name && *name) { + ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar), + &name_utf16le, &name_utf16le_nbytes); + if (ret) + return ret; } - FREE(dentry->file_name); - dentry->file_name = name; - dentry->file_name_nbytes = name_nbytes; - - return dentry_clear_short_name(dentry); + do_dentry_set_name(dentry, name_utf16le, name_utf16le_nbytes); + return 0; } -/* Returns the total length of a WIM alternate data stream entry on-disk, - * including the stream name, the null terminator, AND the padding after the - * entry to align the next ADS entry or dentry on an 8-byte boundary. */ -static u64 -ads_entry_total_length(const struct wim_ads_entry *entry) +/* 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 name_nbytes, u16 short_name_nbytes) { - u64 len = sizeof(struct wim_ads_entry_on_disk); - if (entry->stream_name_nbytes) - len += entry->stream_name_nbytes + 2; - return (len + 7) & ~7; + size_t length = sizeof(struct wim_dentry_on_disk); + if (name_nbytes) + length += (u32)name_nbytes + 2; + if (short_name_nbytes) + length += (u32)short_name_nbytes + 2; + return length; } -/* - * Determine whether to include a "dummy" stream when writing a WIM dentry: - * - * Some versions of Microsoft's WIM software (the boot driver(s) in WinPE 3.0, - * for example) contain a bug where they assume the first alternate data stream - * (ADS) entry of a dentry with a nonzero ADS count specifies the unnamed - * stream, even if it has a name and the unnamed stream is already specified in - * the hash field of the dentry itself. - * - * wimlib has to work around this behavior by carefully emulating the behavior - * of (most versions of) ImageX/WIMGAPI, which move the unnamed stream reference - * into the alternate stream entries whenever there are named data streams, even - * though there is already a field in the dentry itself for the unnamed stream - * reference, which then goes to waste. - */ -static inline bool -inode_needs_dummy_stream(const struct wim_inode *inode) -{ - return (inode->i_num_ads > 0 && - inode->i_num_ads < 0xffff && /* overflow check */ - inode->i_canonical_streams); /* assume the dentry is okay if it - already had an unnamed ADS entry - when it was read in */ -} -/* Calculate the total number of bytes that will be consumed when a WIM dentry - * is written. This includes base dentry and name fields as well as all - * alternate data stream entries and alignment bytes. */ -u64 -dentry_out_total_length(const struct wim_dentry *dentry) +/* 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) { - u64 length = dentry_correct_length_aligned(dentry); - const struct wim_inode *inode = dentry->d_inode; + /* Account for the fixed length portion */ + size_t len = sizeof(struct wim_extra_stream_entry_on_disk); - if (inode_needs_dummy_stream(inode)) - length += ads_entry_total_length(&(struct wim_ads_entry){}); + /* For named streams, account for the variable-length name. */ + if (stream_is_named(strm)) + len += utf16le_len_bytes(strm->stream_name) + 2; - for (u16 i = 0; i < inode->i_num_ads; i++) - length += ads_entry_total_length(&inode->i_ads_entries[i]); - - return length; + /* Account for any necessary padding to the next 8-byte boundary. */ + return ALIGN(len, 8); } -/* Calculate the aligned, total length of a dentry, including all alternate data - * stream entries. Uses dentry->length. */ -static u64 -dentry_in_total_length(const struct wim_dentry *dentry) +/* + * 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. + */ +size_t +dentry_out_total_length(const struct wim_dentry *dentry) { - u64 length = dentry->length; const struct wim_inode *inode = dentry->d_inode; - for (u16 i = 0; i < inode->i_num_ads; i++) - length += ads_entry_total_length(&inode->i_ads_entries[i]); - return (length + 7) & ~7; + size_t len; + + len = dentry_min_len_with_names(dentry->d_name_nbytes, + dentry->d_short_name_nbytes); + len = ALIGN(len, 8); + + len += ALIGN(inode->i_extra_size, 8); + + 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 += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8); + len += ALIGN(sizeof(struct wim_extra_stream_entry_on_disk), 8); + } + } + + return len; } +/* 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) @@ -356,7 +413,7 @@ do_for_dentry_in_tree(struct wim_dentry *dentry, return 0; } - +/* 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) @@ -372,10 +429,25 @@ do_for_dentry_in_tree_depth(struct wim_dentry *dentry, return unlikely((*visitor)(dentry, arg)); } -/* Calls a function on all directory entries in a WIM dentry tree. Logically, - * this is a pre-order traversal (the function is called on a parent dentry - * before its children), but sibling dentries will be visited in order as well. - * */ +/* + * 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 for_dentry_in_tree(struct wim_dentry *root, int (*visitor)(struct wim_dentry *, void *), void *arg) @@ -385,9 +457,10 @@ for_dentry_in_tree(struct wim_dentry *root, return do_for_dentry_in_tree(root, visitor, arg); } -/* Like for_dentry_in_tree(), but the visitor function is always called on a - * dentry's children before on itself. */ -int +/* 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) { @@ -396,97 +469,74 @@ for_dentry_in_tree_depth(struct wim_dentry *root, return do_for_dentry_in_tree_depth(root, visitor, arg); } -/* Calculate the full path of @dentry. */ +/* + * 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->d_full_path'. + * + * Whenever possible, use dentry_full_path() instead of calling this and + * accessing d_full_path directly. + * + * Returns 0 or an error code resulting from a failed string conversion. + */ int calculate_dentry_full_path(struct wim_dentry *dentry) { - tchar *full_path; - u32 full_path_nbytes; - int ret; + size_t ulen; + size_t dummy; + const struct wim_dentry *d; - if (dentry->_full_path) + if (dentry->d_full_path) return 0; - if (dentry_is_root(dentry)) { - static const tchar _root_path[] = {WIM_PATH_SEPARATOR, T('\0')}; - full_path = TSTRDUP(_root_path); - if (full_path == NULL) - return WIMLIB_ERR_NOMEM; - full_path_nbytes = 1 * sizeof(tchar); - } else { - struct wim_dentry *parent; - tchar *parent_full_path; - u32 parent_full_path_nbytes; - size_t filename_nbytes; - - parent = dentry->parent; - if (dentry_is_root(parent)) { - parent_full_path = T(""); - parent_full_path_nbytes = 0; - } else { - if (parent->_full_path == NULL) { - ret = calculate_dentry_full_path(parent); - if (ret) - return ret; - } - parent_full_path = parent->_full_path; - parent_full_path_nbytes = parent->full_path_nbytes; - } + ulen = 0; + d = dentry; + do { + ulen += d->d_name_nbytes / sizeof(utf16lechar); + ulen++; + d = d->d_parent; /* assumes d == d->d_parent for root */ + } while (!dentry_is_root(d)); - /* Append this dentry's name as a tchar string to the full path - * of the parent followed by the path separator */ - #if TCHAR_IS_UTF16LE - filename_nbytes = dentry->file_name_nbytes; - #else - { - int ret = utf16le_to_tstr_nbytes(dentry->file_name, - dentry->file_name_nbytes, - &filename_nbytes); - if (ret) - return ret; - } - #endif + utf16lechar ubuf[ulen]; + utf16lechar *p = &ubuf[ulen]; - full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) + - filename_nbytes; - full_path = MALLOC(full_path_nbytes + sizeof(tchar)); - if (full_path == NULL) - return WIMLIB_ERR_NOMEM; - memcpy(full_path, parent_full_path, parent_full_path_nbytes); - full_path[parent_full_path_nbytes / sizeof(tchar)] = WIM_PATH_SEPARATOR; - #if TCHAR_IS_UTF16LE - memcpy(&full_path[parent_full_path_nbytes / sizeof(tchar) + 1], - dentry->file_name, - filename_nbytes + sizeof(tchar)); - #else - utf16le_to_tstr_buf(dentry->file_name, - dentry->file_name_nbytes, - &full_path[parent_full_path_nbytes / - sizeof(tchar) + 1]); - #endif - } - dentry->_full_path = full_path; - dentry->full_path_nbytes= full_path_nbytes; - return 0; + d = dentry; + do { + p -= d->d_name_nbytes / sizeof(utf16lechar); + memcpy(p, d->d_name, d->d_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->d_full_path, &dummy); } +/* + * 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. + * + * The returned memory will be cached in the dentry, so the caller is not + * responsible for freeing it. + */ tchar * dentry_full_path(struct wim_dentry *dentry) { calculate_dentry_full_path(dentry); - return dentry->_full_path; + return dentry->d_full_path; } static int dentry_calculate_subdir_offset(struct wim_dentry *dentry, void *_subdir_offset_p) { - if (dentry_is_directory(dentry)) { u64 *subdir_offset_p = _subdir_offset_p; struct wim_dentry *child; /* Set offset of directory's child dentries */ - dentry->subdir_offset = *subdir_offset_p; + dentry->d_subdir_offset = *subdir_offset_p; /* Account for child dentries */ for_dentry_child(child, dentry) @@ -495,14 +545,25 @@ dentry_calculate_subdir_offset(struct wim_dentry *dentry, void *_subdir_offset_p /* Account for end-of-directory entry */ *subdir_offset_p += 8; } else { - /* Not a directory; set subdir_offset to 0 */ - dentry->subdir_offset = 0; + /* Not a directory; set the subdir offset to 0 */ + dentry->d_subdir_offset = 0; } return 0; } /* - * Calculates the subdir offsets for a directory tree. + * 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) @@ -515,10 +576,10 @@ static int dentry_compare_names_case_insensitive(const struct wim_dentry *d1, const struct wim_dentry *d2) { - return cmp_utf16le_strings(d1->file_name, - d1->file_name_nbytes / 2, - d2->file_name, - d2->file_name_nbytes / 2, + return cmp_utf16le_strings(d1->d_name, + d1->d_name_nbytes / 2, + d2->d_name, + d2->d_name_nbytes / 2, true); } @@ -527,10 +588,10 @@ static int dentry_compare_names_case_sensitive(const struct wim_dentry *d1, const struct wim_dentry *d2) { - return cmp_utf16le_strings(d1->file_name, - d1->file_name_nbytes / 2, - d2->file_name, - d2->file_name_nbytes / 2, + return cmp_utf16le_strings(d1->d_name, + d1->d_name_nbytes / 2, + d2->d_name, + d2->d_name_nbytes / 2, false); } @@ -557,8 +618,9 @@ _avl_dentry_compare_names(const struct avl_tree_node *n1, } /* Default case sensitivity behavior for searches with - * WIMLIB_CASE_PLATFORM_DEFAULT specified. This can be modified by - * wimlib_global_init(). */ + * 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 @@ -567,8 +629,8 @@ bool default_ignore_case = #endif ; -/* Case-sensitive dentry lookup. Only @file_name and @file_name_nbytes of - * @dummy must be valid. */ +/* Case-sensitive dentry lookup. Only @d_name and @d_name_nbytes of @dummy must + * be valid. */ static struct wim_dentry * dir_lookup(const struct wim_inode *dir, const struct wim_dentry *dummy) { @@ -582,8 +644,8 @@ dir_lookup(const struct wim_inode *dir, const struct wim_dentry *dummy) return avl_tree_entry(node, struct wim_dentry, d_index_node); } -/* Case-insensitive dentry lookup. Only @file_name and @file_name_nbytes of - * @dummy must be valid. */ +/* Case-insensitive dentry lookup. Only @d_name and @d_name_nbytes of @dummy + * must be valid. */ static struct wim_dentry * dir_lookup_ci(const struct wim_inode *dir, const struct wim_dentry *dummy) { @@ -598,8 +660,8 @@ dir_lookup_ci(const struct wim_inode *dir, const struct wim_dentry *dummy) } /* Given a UTF-16LE filename and a directory, look up the dentry for the file. - * Return it if found, otherwise NULL. This is case-sensitive on UNIX and - * case-insensitive on Windows. */ + * 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, @@ -611,8 +673,8 @@ get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry, struct wim_dentry dummy; struct wim_dentry *child; - dummy.file_name = (utf16lechar*)name; - dummy.file_name_nbytes = name_nbytes; + dummy.d_name = (utf16lechar*)name; + dummy.d_name_nbytes = name_nbytes; if (!ignore_case) /* Case-sensitive lookup. */ @@ -654,37 +716,34 @@ get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry, return child; } -/* Returns the child of @dentry that has the file name @name. Returns NULL if - * no child has the name. */ +/* 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 TCHAR_IS_UTF16LE - return get_dentry_child_with_utf16le_name(dentry, name, - tstrlen(name) * sizeof(tchar), - case_type); -#else - utf16lechar *utf16le_name; - size_t utf16le_name_nbytes; int ret; + const utf16lechar *name_utf16le; + size_t name_utf16le_nbytes; struct wim_dentry *child; - ret = tstr_to_utf16le(name, tstrlen(name) * sizeof(tchar), - &utf16le_name, &utf16le_name_nbytes); - if (ret) { - child = NULL; - } else { - child = get_dentry_child_with_utf16le_name(dentry, - utf16le_name, - utf16le_name_nbytes, - case_type); - FREE(utf16le_name); - } + 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; -#endif } +/* 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) @@ -695,7 +754,7 @@ get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path, /* 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_root_dentry(wim); + cur_dentry = wim_get_current_root_dentry(wim); name_start = path; for (;;) { @@ -789,37 +848,31 @@ get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path, struct wim_dentry * get_dentry(WIMStruct *wim, const tchar *path, CASE_SENSITIVITY_TYPE case_type) { -#if TCHAR_IS_UTF16LE - return get_dentry_utf16le(wim, path, case_type); -#else - utf16lechar *path_utf16le; - size_t path_utf16le_nbytes; int ret; + const utf16lechar *path_utf16le; struct wim_dentry *dentry; - ret = tstr_to_utf16le(path, tstrlen(path) * sizeof(tchar), - &path_utf16le, &path_utf16le_nbytes); + ret = tstr_get_utf16le(path, &path_utf16le); if (ret) return NULL; dentry = get_dentry_utf16le(wim, path_utf16le, case_type); - FREE(path_utf16le); + tstr_put_utf16le(path_utf16le); return dentry; -#endif } -/* Takes in a path of length @len in @buf, and transforms it into a string for - * the path of its parent directory. */ +/* 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 *buf, size_t len) +to_parent_name(tchar *path, size_t len) { ssize_t i = (ssize_t)len - 1; - while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR) + while (i >= 0 && path[i] == WIM_PATH_SEPARATOR) i--; - while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR) + while (i >= 0 && path[i] != WIM_PATH_SEPARATOR) i--; - while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR) + while (i >= 0 && path[i] == WIM_PATH_SEPARATOR) i--; - buf[i + 1] = T('\0'); + path[i + 1] = T('\0'); } /* Similar to get_dentry(), but returns the dentry named by @path with the last @@ -838,159 +891,92 @@ get_parent_dentry(WIMStruct *wim, const tchar *path, return get_dentry(wim, buf, case_type); } -#ifdef WITH_FUSE -/* Finds the dentry, lookup table entry, and stream index for a WIM file stream, - * given a path name. +/* + * Create an unlinked dentry. * - * Currently, lookups of this type are only needed if FUSE is enabled. */ -int -wim_pathname_to_stream(WIMStruct *wim, - const tchar *path, - int lookup_flags, - struct wim_dentry **dentry_ret, - struct wim_lookup_table_entry **lte_ret, - u16 *stream_idx_ret) -{ - struct wim_dentry *dentry; - struct wim_lookup_table_entry *lte; - u16 stream_idx; - const tchar *stream_name = NULL; - struct wim_inode *inode; - tchar *p = NULL; - - if (lookup_flags & LOOKUP_FLAG_ADS_OK) { - stream_name = path_stream_name(path); - if (stream_name) { - p = (tchar*)stream_name - 1; - *p = T('\0'); - } - } - - dentry = get_dentry(wim, path, WIMLIB_CASE_SENSITIVE); - if (p) - *p = T(':'); - if (!dentry) - return -errno; - - inode = dentry->d_inode; - - if (!inode->i_resolved) - if (inode_resolve_streams(inode, wim->lookup_table, false)) - return -EIO; - - if (!(lookup_flags & LOOKUP_FLAG_DIRECTORY_OK) - && inode_is_directory(inode)) - return -EISDIR; - - if (stream_name) { - struct wim_ads_entry *ads_entry; - u16 ads_idx; - ads_entry = inode_get_ads_entry(inode, stream_name, - &ads_idx); - if (ads_entry) { - stream_idx = ads_idx + 1; - lte = ads_entry->lte; - goto out; - } else { - return -ENOENT; - } - } else { - lte = inode_unnamed_stream_resolved(inode, &stream_idx); - } -out: - if (dentry_ret) - *dentry_ret = dentry; - if (lte_ret) - *lte_ret = lte; - if (stream_idx_ret) - *stream_idx_ret = stream_idx; - return 0; -} -#endif /* WITH_FUSE */ - -/* Initializations done on every `struct wim_dentry'. */ -static void -dentry_common_init(struct wim_dentry *dentry) -{ - memset(dentry, 0, sizeof(struct wim_dentry)); -} - -/* Creates an unlinked directory entry. */ -int + * @name specifies the long name to give the new dentry. If NULL or empty, the + * new dentry will be given no long name. + * + * 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. + */ +static int new_dentry(const tchar *name, struct wim_dentry **dentry_ret) { struct wim_dentry *dentry; int ret; - dentry = MALLOC(sizeof(struct wim_dentry)); - if (dentry == NULL) + dentry = CALLOC(1, sizeof(struct wim_dentry)); + if (!dentry) return WIMLIB_ERR_NOMEM; - dentry_common_init(dentry); - if (*name) { + if (name && *name) { ret = dentry_set_name(dentry, name); if (ret) { FREE(dentry); - ERROR("Failed to set name on new dentry with name \"%"TS"\"", - name); return ret; } } - dentry->parent = dentry; + dentry->d_parent = dentry; *dentry_ret = dentry; return 0; } -static int -_new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret, - bool timeless) +/* 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 wim_dentry *dentry; + struct wim_inode *inode; int ret; ret = new_dentry(name, &dentry); if (ret) return ret; - if (timeless) - dentry->d_inode = new_timeless_inode(); - else - dentry->d_inode = new_inode(); - if (dentry->d_inode == NULL) { + inode = new_inode(dentry, set_timestamps); + if (!inode) { free_dentry(dentry); return WIMLIB_ERR_NOMEM; } - inode_add_dentry(dentry, dentry->d_inode); *dentry_ret = dentry; return 0; } +/* 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_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret) -{ - return _new_dentry_with_inode(name, dentry_ret, true); -} - -int -new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret) +new_dentry_with_existing_inode(const tchar *name, struct wim_inode *inode, + struct wim_dentry **dentry_ret) { - return _new_dentry_with_inode(name, dentry_ret, false); + int ret = new_dentry(name, dentry_ret); + if (ret) + return ret; + d_associate(*dentry_ret, inode); + inode_ref_blobs(inode); + return 0; } +/* Create an unnamed dentry with a new inode for a directory with the default + * metadata. */ int -new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret) +new_filler_directory(struct wim_dentry **dentry_ret) { int ret; struct wim_dentry *dentry; - DEBUG("Creating filler directory \"%"TS"\"", name); - ret = new_dentry_with_inode(name, &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_resolved = 1; dentry->d_inode->i_attributes = FILE_ATTRIBUTE_DIRECTORY; *dentry_ret = dentry; return 0; @@ -1009,63 +995,68 @@ dentry_tree_clear_inode_visited(struct wim_dentry *root) for_dentry_in_tree(root, dentry_clear_inode_visited, NULL); } -/* Frees a WIM dentry. +/* + * Free a WIM dentry. * - * The corresponding inode (if any) 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. + */ void free_dentry(struct wim_dentry *dentry) { if (dentry) { - FREE(dentry->file_name); - FREE(dentry->short_name); - FREE(dentry->_full_path); - if (dentry->d_inode) - put_inode(dentry->d_inode); + d_disassociate(dentry); + FREE(dentry->d_name); + FREE(dentry->d_short_name); + FREE(dentry->d_full_path); FREE(dentry); } } -/* This function is passed as an argument to for_dentry_in_tree_depth() in order - * to free a directory tree. */ static int -do_free_dentry(struct wim_dentry *dentry, void *_lookup_table) +do_free_dentry(struct wim_dentry *dentry, void *_ignore) { - struct wim_lookup_table *lookup_table = _lookup_table; - - if (lookup_table) { - struct wim_inode *inode = dentry->d_inode; - for (unsigned i = 0; i <= inode->i_num_ads; i++) { - struct wim_lookup_table_entry *lte; + free_dentry(dentry); + return 0; +} - lte = inode_stream_lte(inode, i, lookup_table); - if (lte) - lte_decrement_refcnt(lte, lookup_table); - } - } +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; } /* - * Recursively frees all directory entries in the specified tree. + * Free all dentries in a tree. * * @root: - * The root of the tree. + * The root of the dentry tree to free. If NULL, this function has no + * effect. * - * @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. + * @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. * - * This also puts references to the corresponding inodes. + * This function also releases references to the corresponding inodes. * - * This does *not* unlink @root from its parent directory (if it has one). + * 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. */ void -free_dentry_tree(struct wim_dentry *root, struct wim_lookup_table *lookup_table) +free_dentry_tree(struct wim_dentry *root, struct blob_table *blob_table) { - for_dentry_in_tree_depth(root, do_free_dentry, lookup_table); + 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); } /* Insert the @child dentry into the case sensitive index of the @dir directory. @@ -1100,21 +1091,21 @@ dir_index_child_ci(struct wim_inode *dir, struct wim_dentry *child) return avl_tree_entry(duplicate, struct wim_dentry, d_index_node_ci); } -/* Removes the specified dentry from its directory's case-sensitive index. */ +/* Remove the specified dentry from its directory's case-sensitive index. */ static void dir_unindex_child(struct wim_inode *dir, struct wim_dentry *child) { avl_tree_remove(&dir->i_children, &child->d_index_node); } -/* Removes the specified dentry from its directory's case-insensitive index. */ +/* 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) { avl_tree_remove(&dir->i_children_ci, &child->d_index_node_ci); } -/* Returns true iff the specified dentry is in its parent directory's +/* 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) @@ -1123,10 +1114,13 @@ dentry_in_ci_index(const struct wim_dentry *dentry) } /* - * Links a dentry into the directory tree. + * Link a dentry into the tree. + * + * @parent: + * The dentry that will be the parent of @child. It must name a directory. * - * @parent: The dentry that will be the parent of @child. - * @child: The dentry to link. + * @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 @@ -1155,20 +1149,23 @@ dentry_add_child(struct wim_dentry *parent, struct wim_dentry *child) } else { INIT_LIST_HEAD(&child->d_ci_conflict_list); } - child->parent = parent; + child->d_parent = parent; return NULL; } -/* Unlink a WIM dentry from the directory entry tree. */ +/* Unlink a dentry from the tree. */ void unlink_dentry(struct wim_dentry *dentry) { struct wim_inode *dir; - if (dentry_is_root(dentry)) + /* 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; - dir = dentry->parent->d_inode; + dir = dentry->d_parent->d_inode; dir_unindex_child(dir, dentry); @@ -1178,7 +1175,7 @@ unlink_dentry(struct wim_dentry *dentry) if (!list_empty(&dentry->d_ci_conflict_list)) { /* Make a different case-insensitively-the-same dentry - * be the "representative" in the search index. */ + * be the "representative" in the search index. */ struct list_head *next; struct wim_dentry *other; struct wim_dentry *existing; @@ -1190,21 +1187,210 @@ unlink_dentry(struct wim_dentry *dentry) } } 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; +} + +static int +read_extra_data(const u8 *p, const u8 *end, struct wim_inode *inode) +{ + while (((uintptr_t)p & 7) && p < end) + p++; + + 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; +} + +/* + * 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) +{ + 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; + } + } +} + +/* + * 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) +{ + 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; + } + } + + if (!found_unnamed_data_stream && unnamed_stream_with_zero_hash != NULL) + unnamed_stream_with_zero_hash->stream_type = STREAM_TYPE_DATA; +} + +/* + * Read and interpret the collection of streams for the specified inode. + */ +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) +{ + const u8 *orig_p = p; + + 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; + } + + /* 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; + u16 name_nbytes; + + strm = &inode->i_streams[i]; + + strm->stream_id = i; + + /* 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; + + disk_strm = (const struct wim_extra_stream_entry_on_disk *)p; + + /* Read the length field */ + length = ALIGN(le64_to_cpu(disk_strm->length), 8); + + /* 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; + + /* Read the rest of the fixed-length data. */ + + copy_hash(strm->_stream_hash, disk_strm->hash); + name_nbytes = le16_to_cpu(disk_strm->name_nbytes); + + /* 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; + + /* 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; + } + + strm->stream_type = STREAM_TYPE_UNKNOWN; + + p += length; + } + + 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; } -/* Reads a WIM directory entry, including all alternate data stream entries that - * follow it, from the WIM image's metadata resource. */ +/* 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, struct wim_dentry **dentry_ret) + 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; + u16 name_nbytes; u64 calculated_size; int ret; @@ -1217,22 +1403,14 @@ read_dentry(const u8 * restrict buf, size_t buf_len, /* Check for buffer overrun. */ if (unlikely(offset + sizeof(u64) > buf_len || offset + sizeof(u64) < offset)) - { - ERROR("Directory entry starting at %"PRIu64" ends past the " - "end of the metadata resource (size %zu)", - offset, buf_len); return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - } /* Get pointer to the dentry data. */ p = &buf[offset]; disk_dentry = (const struct wim_dentry_on_disk*)p; - if (unlikely((uintptr_t)p & 7)) - WARNING("WIM dentry is not 8-byte aligned"); - /* Get dentry length. */ - length = le64_to_cpu(disk_dentry->length); + length = ALIGN(le64_to_cpu(disk_dentry->length), 8); /* Check for end-of-directory. */ if (length <= 8) { @@ -1241,83 +1419,57 @@ read_dentry(const u8 * restrict buf, size_t buf_len, } /* Validate dentry length. */ - if (unlikely(length < sizeof(struct wim_dentry_on_disk))) { - ERROR("Directory entry has invalid length of %"PRIu64" bytes", - length); + if (unlikely(length < sizeof(struct wim_dentry_on_disk))) return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - } /* Check for buffer overrun. */ if (unlikely(offset + length > buf_len || offset + length < offset)) - { - ERROR("Directory entry at offset %"PRIu64" and with size " - "%"PRIu64" ends past the end of the metadata resource " - "(size %zu)", offset, length, buf_len); return WIMLIB_ERR_INVALID_METADATA_RESOURCE; - } /* Allocate new dentry structure, along with a preliminary inode. */ - ret = new_dentry_with_timeless_inode(T(""), &dentry); + ret = new_dentry_with_new_inode(NULL, false, &dentry); if (ret) return ret; - dentry->length = length; inode = dentry->d_inode; /* 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); + dentry->d_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); - copy_hash(inode->i_hash, disk_dentry->unnamed_stream_hash); + inode->i_unknown_0x54 = le32_to_cpu(disk_dentry->unknown_0x54); - /* 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. 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 that this means the WIM file - * cannot archive hard-linked reparse points. Such a thing - * doesn't really make sense anyway, although I believe it's - * theoretically possible to have them on NTFS. */ + inode->i_rp_reserved = le16_to_cpu(disk_dentry->reparse.rp_reserved); + inode->i_rp_flags = le16_to_cpu(disk_dentry->reparse.rp_flags); + /* Leave inode->i_ino at 0. Note: this means that WIM cannot + * represent multiple hard links to a reparse point file. */ } else { - 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); } - inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams); /* Now onto reading the names. There are two of them: the (long) file * name, and the short name. */ short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes); - file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes); + name_nbytes = le16_to_cpu(disk_dentry->name_nbytes); - if (unlikely((short_name_nbytes & 1) | (file_name_nbytes & 1))) { - ERROR("Dentry name is not valid UTF-16 (odd number of bytes)!"); + if (unlikely((short_name_nbytes & 1) | (name_nbytes & 1))) { ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; goto err_free_dentry; } /* 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_nbytes, - short_name_nbytes); - - if (unlikely(dentry->length < calculated_size)) { - ERROR("Unexpected end of directory entry! (Expected " - "at least %"PRIu64" bytes, got %"PRIu64" bytes.)", - calculated_size, dentry->length); + * the length of the dentry is large enough to actually hold them. */ + calculated_size = dentry_min_len_with_names(name_nbytes, + short_name_nbytes); + + if (unlikely(length < calculated_size)) { ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; goto err_free_dentry; } @@ -1327,56 +1479,47 @@ read_dentry(const u8 * restrict buf, size_t buf_len, /* 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 = MALLOC(file_name_nbytes + 2); - if (dentry->file_name == NULL) { + if (name_nbytes) { + dentry->d_name = utf16le_dupz(p, name_nbytes); + if (unlikely(!dentry->d_name)) { ret = WIMLIB_ERR_NOMEM; goto err_free_dentry; } - dentry->file_name_nbytes = file_name_nbytes; - memcpy(dentry->file_name, p, file_name_nbytes); - p += file_name_nbytes + 2; - dentry->file_name[file_name_nbytes / 2] = cpu_to_le16(0); + dentry->d_name_nbytes = name_nbytes; + p += (u32)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_nbytes) { - dentry->short_name = MALLOC(short_name_nbytes + 2); - if (dentry->short_name == NULL) { + dentry->d_short_name = utf16le_dupz(p, short_name_nbytes); + if (unlikely(!dentry->d_short_name)) { ret = WIMLIB_ERR_NOMEM; goto err_free_dentry; } - dentry->short_name_nbytes = short_name_nbytes; - memcpy(dentry->short_name, p, short_name_nbytes); - p += short_name_nbytes + 2; - dentry->short_name[short_name_nbytes / 2] = cpu_to_le16(0); + dentry->d_short_name_nbytes = short_name_nbytes; + p += (u32)short_name_nbytes + 2; } - /* Align the dentry length. */ - dentry->length = (dentry->length + 7) & ~7; + /* 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; - /* Read the alternate data streams, if present. inode->i_num_ads tells - * us how many they are, and they will directly follow the dentry in the - * metadata resource buffer. - * - * Note that each alternate data stream entry begins on an 8-byte - * aligned boundary, and the alternate data stream entries seem to NOT - * be included in the dentry->length field for some reason. */ - if (unlikely(inode->i_num_ads != 0)) { - ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; - if (offset + dentry->length > buf_len || - (ret = read_ads_entries(&buf[offset + dentry->length], - inode, - buf_len - offset - dentry->length))) - { - ERROR("Failed to read alternate data stream " - "entries of WIM dentry \"%"WS"\"", - dentry->file_name); - 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; @@ -1385,28 +1528,17 @@ err_free_dentry: return ret; } -static const tchar * -dentry_get_file_type_string(const struct wim_dentry *dentry) -{ - const struct wim_inode *inode = dentry->d_inode; - if (inode_is_directory(inode)) - return T("directory"); - else if (inode_is_symlink(inode)) - return T("symbolic link"); - else - return T("file"); -} - +/* Is the dentry named "." or ".." ? */ static bool dentry_is_dot_or_dotdot(const struct wim_dentry *dentry) { - 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('.')) + if (dentry->d_name_nbytes <= 4) { + if (dentry->d_name_nbytes == 4) { + if (dentry->d_name[0] == cpu_to_le16('.') && + dentry->d_name[1] == cpu_to_le16('.')) return true; - } else if (dentry->file_name_nbytes == 2) { - if (dentry->file_name[0] == cpu_to_le16('.')) + } else if (dentry->d_name_nbytes == 2) { + if (dentry->d_name[0] == cpu_to_le16('.')) return true; } } @@ -1417,14 +1549,14 @@ 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; + u64 cur_offset = dir->d_subdir_offset; /* Check for cyclic directory structure, which would cause infinite * recursion if not handled. */ - for (struct wim_dentry *d = dir->parent; - !dentry_is_root(d); d = d->parent) + for (struct wim_dentry *d = dir->d_parent; + !dentry_is_root(d); d = d->d_parent) { - if (unlikely(d->subdir_offset == cur_offset)) { + if (unlikely(d->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)); @@ -1438,7 +1570,7 @@ read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len, int ret; /* Read next child of @dir. */ - ret = read_dentry(buf, buf_len, cur_offset, &child); + ret = read_dentry(buf, buf_len, &cur_offset, &child); if (ret) return ret; @@ -1446,13 +1578,6 @@ read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len, if (child == NULL) return 0; - /* 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_in_total_length(child); - /* All dentries except the root should be named. */ if (unlikely(!dentry_has_long_name(child))) { WARNING("Ignoring unnamed dentry in " @@ -1475,21 +1600,17 @@ read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len, /* We already found a dentry with this same * case-sensitive long name. Only keep the first one. */ - const tchar *child_type, *duplicate_type; - child_type = dentry_get_file_type_string(child); - duplicate_type = dentry_get_file_type_string(duplicate); - WARNING("Ignoring duplicate %"TS" \"%"TS"\" " - "(the WIM image already contains a %"TS" " + WARNING("Ignoring duplicate file \"%"TS"\" " + "(the WIM image already contains a file " "at that path with the exact same name)", - child_type, dentry_full_path(duplicate), - duplicate_type); + 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 (child->d_subdir_offset != 0) { if (likely(dentry_is_directory(child))) { ret = read_dentry_tree_recursive(buf, buf_len, @@ -1506,7 +1627,7 @@ read_dentry_tree_recursive(const u8 * restrict buf, size_t buf_len, } /* - * Read a tree of dentries (directory entries) from a WIM metadata resource. + * Read a tree of dentries from a WIM metadata resource. * * @buf: * Buffer containing an uncompressed WIM metadata resource. @@ -1536,7 +1657,7 @@ read_dentry_tree(const u8 *buf, size_t buf_len, DEBUG("Reading dentry tree (root_offset=%"PRIu64")", root_offset); - ret = read_dentry(buf, buf_len, root_offset, &root); + ret = read_dentry(buf, buf_len, &root_offset, &root); if (ret) return ret; @@ -1546,12 +1667,7 @@ read_dentry_tree(const u8 *buf, size_t buf_len, { WARNING("The root directory has a nonempty name; " "removing it."); - FREE(root->file_name); - FREE(root->short_name); - root->file_name = NULL; - root->short_name = NULL; - root->file_name_nbytes = 0; - root->short_name_nbytes = 0; + dentry_set_name(root, NULL); } if (unlikely(!dentry_is_directory(root))) { @@ -1560,7 +1676,7 @@ read_dentry_tree(const u8 *buf, size_t buf_len, goto err_free_dentry_tree; } - if (likely(root->subdir_offset != 0)) { + if (likely(root->d_subdir_offset != 0)) { ret = read_dentry_tree_recursive(buf, buf_len, root); if (ret) goto err_free_dentry_tree; @@ -1577,46 +1693,45 @@ err_free_dentry_tree: return ret; } -/* - * Writes a WIM alternate data stream (ADS) entry to an output buffer. - * - * @ads_entry: The ADS entry structure. - * @hash: The hash field to use (instead of the one in the ADS entry). - * @p: The memory location to write the data to. - * - * Returns a pointer to the byte after the last byte written. - */ static u8 * -write_ads_entry(const struct wim_ads_entry *ads_entry, - const u8 *hash, u8 * restrict p) +write_extra_stream_entry(u8 * restrict p, const utf16lechar * restrict name, + const u8 * restrict hash) { - struct wim_ads_entry_on_disk *disk_ads_entry = - (struct wim_ads_entry_on_disk*)p; + struct wim_extra_stream_entry_on_disk *disk_strm = + (struct wim_extra_stream_entry_on_disk *)p; u8 *orig_p = p; + size_t name_nbytes; - disk_ads_entry->reserved = cpu_to_le64(ads_entry->reserved); - copy_hash(disk_ads_entry->hash, hash); - disk_ads_entry->stream_name_nbytes = cpu_to_le16(ads_entry->stream_name_nbytes); - p += sizeof(struct wim_ads_entry_on_disk); - if (ads_entry->stream_name_nbytes) { - p = mempcpy(p, ads_entry->stream_name, - ads_entry->stream_name_nbytes + 2); - } + 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 */ while ((uintptr_t)p & 7) *p++ = 0; - disk_ads_entry->length = cpu_to_le64(p - orig_p); + disk_strm->length = cpu_to_le64(p - orig_p); return p; } /* - * Writes a WIM dentry to an output buffer. + * 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. * - * @dentry: The dentry structure. - * @p: The memory location to write the data to. + * @p: + * The memory location to which to write the data. * - * Returns the pointer to the byte after the last byte we wrote as part of the - * dentry, including any alternate data stream entries. + * Returns a pointer to the byte following the last written. */ static u8 * write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p) @@ -1624,76 +1739,128 @@ write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p) const struct wim_inode *inode; struct wim_dentry_on_disk *disk_dentry; const u8 *orig_p; - const u8 *hash; - bool use_dummy_stream; - u16 num_ads; wimlib_assert(((uintptr_t)p & 7) == 0); /* 8 byte aligned */ orig_p = p; inode = dentry->d_inode; - use_dummy_stream = inode_needs_dummy_stream(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->subdir_offset = cpu_to_le64(dentry->d_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 (use_dummy_stream) - hash = zero_hash; - else - hash = inode_stream_hash(inode, 0); - copy_hash(disk_dentry->unnamed_stream_hash, hash); + disk_dentry->unknown_0x54 = cpu_to_le32(inode->i_unknown_0x54); 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); + disk_dentry->reparse.rp_reserved = cpu_to_le16(inode->i_rp_reserved); + disk_dentry->reparse.rp_flags = cpu_to_le16(inode->i_rp_flags); } 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); } - num_ads = inode->i_num_ads; - if (use_dummy_stream) - num_ads++; - disk_dentry->num_alternate_data_streams = cpu_to_le16(num_ads); - disk_dentry->short_name_nbytes = cpu_to_le16(dentry->short_name_nbytes); - disk_dentry->file_name_nbytes = cpu_to_le16(dentry->file_name_nbytes); + + disk_dentry->short_name_nbytes = cpu_to_le16(dentry->d_short_name_nbytes); + disk_dentry->name_nbytes = cpu_to_le16(dentry->d_name_nbytes); p += sizeof(struct wim_dentry_on_disk); wimlib_assert(dentry_is_root(dentry) != dentry_has_long_name(dentry)); if (dentry_has_long_name(dentry)) - p = mempcpy(p, dentry->file_name, dentry->file_name_nbytes + 2); + p = mempcpy(p, dentry->d_name, (u32)dentry->d_name_nbytes + 2); if (dentry_has_short_name(dentry)) - p = mempcpy(p, dentry->short_name, dentry->short_name_nbytes + 2); + p = mempcpy(p, dentry->d_short_name, (u32)dentry->d_short_name_nbytes + 2); /* Align to 8-byte boundary */ while ((uintptr_t)p & 7) *p++ = 0; - /* 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...). Furthermore, the dentry may - * have been renamed, thus changing its needed length. */ - disk_dentry->length = cpu_to_le64(p - orig_p); + if (inode->i_extra_size) { + /* Extra tagged items --- not usually present. */ + p = mempcpy(p, inode->i_extra, inode->i_extra_size); - if (use_dummy_stream) { - hash = inode_unnamed_stream_hash(inode); - p = write_ads_entry(&(struct wim_ads_entry){}, hash, p); + /* Align to 8-byte boundary */ + while ((uintptr_t)p & 7) + *p++ = 0; } - /* Write the alternate data streams entries, if any. */ - for (u16 i = 0; i < inode->i_num_ads; i++) { - hash = inode_stream_hash(inode, i + 1); - p = write_ads_entry(&inode->i_ads_entries[i], hash, p); + disk_dentry->length = cpu_to_le64(p - orig_p); + + /* Streams */ + + if (unlikely(inode->i_attributes & FILE_ATTRIBUTE_ENCRYPTED)) { + const struct wim_inode_stream *efs_strm; + const u8 *efs_hash; + + efs_strm = inode_get_unnamed_stream(inode, STREAM_TYPE_EFSRPC_RAW_DATA); + 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; + 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 (unlikely(have_reparse_point_stream || have_named_data_stream)) { + + unsigned num_extra_streams = 0; + + 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++; + + 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++; + } + } + wimlib_assert(num_extra_streams <= 0xFFFF); + + disk_dentry->num_extra_streams = cpu_to_le16(num_extra_streams); + } else { + copy_hash(disk_dentry->default_hash, unnamed_data_stream_hash); + disk_dentry->num_extra_streams = cpu_to_le16(0); + } } return p; @@ -1702,7 +1869,7 @@ write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p) static int write_dir_dentries(struct wim_dentry *dir, void *_pp) { - if (dir->subdir_offset != 0) { + if (dir->d_subdir_offset != 0) { u8 **pp = _pp; u8 *p = *pp; struct wim_dentry *child; @@ -1719,18 +1886,26 @@ write_dir_dentries(struct wim_dentry *dir, void *_pp) return 0; } -/* Writes a directory tree to the metadata resource. +/* + * Write a directory tree to the metadata resource. + * + * @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. * - * @root: Root of the dentry tree. - * @p: Pointer to a buffer with enough space for the dentry tree. + * @p: + * Pointer to a buffer with enough space for the dentry tree. This size + * must have been obtained by calculate_subdir_offsets(). * - * Returns pointer to the byte after the last byte we wrote. + * Returns a pointer to the byte following the last written. */ u8 * write_dentry_tree(struct wim_dentry *root, u8 *p) { DEBUG("Writing dentry tree."); - wimlib_assert(dentry_is_root(root)); + + wimlib_assert(root != NULL); /* write root dentry and end-of-directory entry following it */ p = write_dentry(root, p);