X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Fdentry.c;h=af363c45187794f0a382c33f29965dd3b4ea555b;hp=8f591c3d9f57c18f63d89ae90c72504a895c0608;hb=0ddd8e57f633908c771b606cc606f434c15ac832;hpb=594502d26951aa89e374599de862486c1e7a6878 diff --git a/src/dentry.c b/src/dentry.c index 8f591c3d..af363c45 100644 --- a/src/dentry.c +++ b/src/dentry.c @@ -26,20 +26,203 @@ * wimlib; if not, see http://www.gnu.org/licenses/. */ -#include "buffer_io.h" -#include "dentry.h" -#include "lookup_table.h" -#include "timestamp.h" -#include "wimlib_internal.h" +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "wimlib.h" +#include "wimlib/dentry.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 +/* WIM alternate data stream entry (on-disk format) */ +struct wim_ads_entry_on_disk { + /* Length of the entry, in bytes. This apparently includes all + * fixed-length fields, plus the stream name and null terminator if + * present, and the padding up to an 8 byte boundary. wimlib is a + * little less strict when reading the entries, and only requires that + * the number of bytes from this field is at least as large as the size + * of the fixed length fields and stream name without null terminator. + * */ + le64 length; + + le64 reserved; + + /* SHA1 message digest of the uncompressed stream; or, alternatively, + * can be all zeroes if the stream has zero length. */ + u8 hash[SHA1_HASH_SIZE]; + + /* Length of the stream name, in bytes. 0 if the stream is unnamed. */ + le16 stream_name_nbytes; + + /* Stream name in UTF-16LE. It is @stream_name_nbytes bytes long, + * excluding the the null terminator. There is a null terminator + * character if @stream_name_nbytes != 0; i.e., if this stream is named. + * */ + utf16lechar stream_name[]; +} _packed_attribute; + +#define WIM_ADS_ENTRY_DISK_SIZE 38 + +/* 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. + * + * 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. */ + le64 length; + + /* Attributes of 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; + + /* 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.) + * + * 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. + * + * 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. + */ + u8 unnamed_stream_hash[SHA1_HASH_SIZE]; + + /* 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. 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. + */ + 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 alternate data stream entries that directly follow this + * dentry on-disk. */ + le16 num_alternate_data_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. */ + 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; + + /* Followed by variable length file name, in UTF16-LE, if + * file_name_nbytes != 0. Includes null terminator. */ + /*utf16lechar file_name[];*/ + + /* Followed by variable length short name, in UTF16-LE, if + * short_name_nbytes != 0. Includes null terminator. */ + /*utf16lechar short_name[];*/ +} _packed_attribute; + +#define WIM_DENTRY_DISK_SIZE 102 + /* 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) +dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes) { - u64 length = WIM_DENTRY_DISK_SIZE; + u64 length = sizeof(struct wim_dentry_on_disk); if (file_name_nbytes) length += file_name_nbytes + 2; if (short_name_nbytes) @@ -52,28 +235,13 @@ __dentry_correct_length_unaligned(u16 file_name_nbytes, u16 short_name_nbytes) * ignored; also, this excludes any alternate data stream entries that may * follow the dentry. */ static u64 -dentry_correct_length_unaligned(const struct wim_dentry *dentry) -{ - return __dentry_correct_length_unaligned(dentry->file_name_nbytes, - dentry->short_name_nbytes); -} - -/* Return the "correct" value to write in the length field of a WIM dentry, - * based on the file name length and short name length. */ -static u64 -dentry_correct_length(const struct wim_dentry *dentry) +dentry_correct_length_aligned(const struct wim_dentry *dentry) { - return (dentry_correct_length_unaligned(dentry) + 7) & ~7; -} + u64 len; -/* Return %true iff the alternate data stream entry @entry has the UTF-16LE - * stream name @name that has length @name_nbytes bytes. */ -static inline bool -ads_entry_has_name(const struct wim_ads_entry *entry, - const utf16lechar *name, size_t name_nbytes) -{ - return entry->stream_name_nbytes == name_nbytes && - memcmp(entry->stream_name, name, name_nbytes) == 0; + len = dentry_correct_length_unaligned(dentry->file_name_nbytes, + dentry->short_name_nbytes); + return (len + 7) & ~7; } /* Duplicates a string of system-dependent encoding into a UTF-16LE string and @@ -89,7 +257,7 @@ get_utf16le_name(const tchar *name, utf16lechar **name_utf16le_ret, #if TCHAR_IS_UTF16LE name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar); name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar)); - if (!name_utf16le) + if (name_utf16le == NULL) return WIMLIB_ERR_NOMEM; memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar)); ret = 0; @@ -127,7 +295,6 @@ set_dentry_name(struct wim_dentry *dentry, const tchar *new_name) dentry->short_name = NULL; dentry->short_name_nbytes = 0; } - dentry->length = dentry_correct_length(dentry); } return ret; } @@ -138,37 +305,64 @@ set_dentry_name(struct wim_dentry *dentry, const tchar *new_name) static u64 ads_entry_total_length(const struct wim_ads_entry *entry) { - u64 len = WIM_ADS_ENTRY_DISK_SIZE; + u64 len = sizeof(struct wim_ads_entry_on_disk); if (entry->stream_name_nbytes) len += entry->stream_name_nbytes + 2; return (len + 7) & ~7; } +/* + * 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 */ +} -static u64 -__dentry_total_length(const struct wim_dentry *dentry, u64 length) +/* 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) { + u64 length = dentry_correct_length_aligned(dentry); const struct wim_inode *inode = dentry->d_inode; + + if (inode_needs_dummy_stream(inode)) + length += ads_entry_total_length(&(struct wim_ads_entry){}); + for (u16 i = 0; i < inode->i_num_ads; i++) length += ads_entry_total_length(&inode->i_ads_entries[i]); - return (length + 7) & ~7; -} -/* Calculate the aligned *total* length of an on-disk WIM dentry. This includes - * all alternate data streams. */ -u64 -dentry_correct_total_length(const struct wim_dentry *dentry) -{ - return __dentry_total_length(dentry, - dentry_correct_length_unaligned(dentry)); + return length; } -/* Like dentry_correct_total_length(), but use the existing dentry->length field - * instead of calculating its "correct" value. */ +/* Calculate the aligned, total length of a dentry, including all alternate data + * stream entries. Uses dentry->length. */ static u64 -dentry_total_length(const struct wim_dentry *dentry) +dentry_in_total_length(const struct wim_dentry *dentry) { - return __dentry_total_length(dentry, dentry->length); + 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; } int @@ -230,18 +424,36 @@ for_dentry_tree_in_rbtree(struct rb_node *node, int ret; if (node) { ret = for_dentry_tree_in_rbtree(node->rb_left, visitor, arg); - if (ret != 0) + if (ret) return ret; ret = for_dentry_in_tree(rbnode_dentry(node), visitor, arg); - if (ret != 0) + if (ret) return ret; ret = for_dentry_tree_in_rbtree(node->rb_right, visitor, arg); - if (ret != 0) + if (ret) return ret; } return 0; } +/* + * Iterate over all children of @dentry, calling the function @visitor, passing + * it a child dentry and the extra argument @arg. + * + * Note: this function iterates over ALL child dentries, even those with the + * same case-insensitive name. + * + * Note: this function clobbers the tmp_list field of the child dentries. */ +int +for_dentry_child(const struct wim_dentry *dentry, + int (*visitor)(struct wim_dentry *, void *), + void *arg) +{ + return for_dentry_in_rbtree(dentry->d_inode->i_children.rb_node, + visitor, + 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. @@ -250,13 +462,16 @@ int for_dentry_in_tree(struct wim_dentry *root, int (*visitor)(struct wim_dentry*, void*), void *arg) { - int ret = visitor(root, arg); - if (ret == 0) { - ret = for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node, - visitor, - arg); - } - return ret; + int ret; + + if (root == NULL) + return 0; + ret = (*visitor)(root, arg); + if (ret) + return ret; + return for_dentry_tree_in_rbtree(root->d_inode->i_children.rb_node, + visitor, + arg); } /* Like for_dentry_in_tree(), but the visitor function is always called on a @@ -266,16 +481,19 @@ for_dentry_in_tree_depth(struct wim_dentry *root, int (*visitor)(struct wim_dentry*, void*), void *arg) { int ret; + + if (root == NULL) + return 0; ret = for_dentry_tree_in_rbtree_depth(root->d_inode->i_children.rb_node, visitor, arg); - if (ret == 0) - ret = visitor(root, arg); - return ret; + if (ret) + return ret; + return (*visitor)(root, arg); } /* Calculate the full path of @dentry. The full path of its parent must have * already been calculated, or it must be the root dentry. */ -static int +int calculate_dentry_full_path(struct wim_dentry *dentry) { tchar *full_path; @@ -286,8 +504,9 @@ calculate_dentry_full_path(struct wim_dentry *dentry) return 0; if (dentry_is_root(dentry)) { - full_path = TSTRDUP(T("/")); - if (!full_path) + 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 { @@ -301,7 +520,7 @@ calculate_dentry_full_path(struct wim_dentry *dentry) parent_full_path = T(""); parent_full_path_nbytes = 0; } else { - if (!parent->_full_path) { + if (parent->_full_path == NULL) { ret = calculate_dentry_full_path(parent); if (ret) return ret; @@ -327,10 +546,10 @@ calculate_dentry_full_path(struct wim_dentry *dentry) full_path_nbytes = parent_full_path_nbytes + sizeof(tchar) + filename_nbytes; full_path = MALLOC(full_path_nbytes + sizeof(tchar)); - if (!full_path) + 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)] = T('/'); + 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, @@ -369,7 +588,7 @@ dentry_full_path(struct wim_dentry *dentry) static int increment_subdir_offset(struct wim_dentry *dentry, void *subdir_offset_p) { - *(u64*)subdir_offset_p += dentry_correct_total_length(dentry); + *(u64*)subdir_offset_p += dentry_out_total_length(dentry); return 0; } @@ -417,42 +636,133 @@ calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p) } static int -compare_utf16le_names(const utf16lechar *name1, size_t nbytes1, - const utf16lechar *name2, size_t nbytes2) +dentry_compare_names_case_insensitive(const struct wim_dentry *d1, + const struct wim_dentry *d2) { - int result = memcmp(name1, name2, min(nbytes1, nbytes2)); - if (result) - return result; - else - return (int)nbytes1 - (int)nbytes2; + return cmp_utf16le_strings(d1->file_name, + d1->file_name_nbytes / 2, + d2->file_name, + d2->file_name_nbytes / 2, + true); } static int -dentry_compare_names(const struct wim_dentry *d1, const struct wim_dentry *d2) +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, + false); +} + +/* Return %true iff the alternate data stream entry @entry has the UTF-16LE + * stream name @name that has length @name_nbytes bytes. */ +static inline bool +ads_entry_has_name(const struct wim_ads_entry *entry, + const utf16lechar *name, size_t name_nbytes, + bool ignore_case) +{ + return 0 == cmp_utf16le_strings(name, + name_nbytes / 2, + entry->stream_name, + entry->stream_name_nbytes / 2, + ignore_case); +} + +/* Default case sensitivity behavior for searches with + * WIMLIB_CASE_PLATFORM_DEFAULT specified. This can be modified by + * wimlib_global_init(). */ +bool default_ignore_case = +#ifdef __WIN32__ + true +#else + false +#endif +; + +static bool +will_ignore_case(CASE_SENSITIVITY_TYPE case_type) { - return compare_utf16le_names(d1->file_name, d1->file_name_nbytes, - d2->file_name, d2->file_name_nbytes); + if (case_type == WIMLIB_CASE_SENSITIVE) + return false; + if (case_type == WIMLIB_CASE_INSENSITIVE) + return true; + + return default_ignore_case; } +/* 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. */ struct wim_dentry * get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry, const utf16lechar *name, - size_t name_nbytes) + size_t name_nbytes, + CASE_SENSITIVITY_TYPE case_ctype) { - struct rb_node *node = dentry->d_inode->i_children.rb_node; + struct rb_node *node; + + bool ignore_case = will_ignore_case(case_ctype); + + if (ignore_case) + node = dentry->d_inode->i_children_case_insensitive.rb_node; + else + node = dentry->d_inode->i_children.rb_node; + struct wim_dentry *child; while (node) { - child = rbnode_dentry(node); - int result = compare_utf16le_names(name, name_nbytes, - child->file_name, - child->file_name_nbytes); - if (result < 0) + if (ignore_case) + child = rb_entry(node, struct wim_dentry, rb_node_case_insensitive); + else + child = rb_entry(node, struct wim_dentry, rb_node); + + int result = cmp_utf16le_strings(name, + name_nbytes / 2, + child->file_name, + child->file_name_nbytes / 2, + ignore_case); + if (result < 0) { node = node->rb_left; - else if (result > 0) + } else if (result > 0) { node = node->rb_right; - else + } else if (!ignore_case || + list_empty(&child->case_insensitive_conflict_list)) { + return child; + } else { + /* Multiple dentries have the same case-insensitive + * name, and a case-insensitive lookup is being + * performed. Choose the dentry with the same + * case-sensitive name, if one exists; otherwise print a + * warning and choose one arbitrarily. */ + struct wim_dentry *alt = child; + size_t num_alts = 0; + + do { + num_alts++; + if (0 == cmp_utf16le_strings(name, + name_nbytes / 2, + alt->file_name, + alt->file_name_nbytes / 2, + false)) + return alt; + alt = list_entry(alt->case_insensitive_conflict_list.next, + struct wim_dentry, + case_insensitive_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->case_insensitive_conflict_list.next, + struct wim_dentry, + case_insensitive_conflict_list))); return child; + } } return NULL; } @@ -460,11 +770,13 @@ get_dentry_child_with_utf16le_name(const struct wim_dentry *dentry, /* Returns the child of @dentry that has the file name @name. Returns NULL if * no child has the name. */ struct wim_dentry * -get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name) +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)); + tstrlen(name) * sizeof(tchar), + case_type); #else utf16lechar *utf16le_name; size_t utf16le_name_nbytes; @@ -478,7 +790,8 @@ get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name) } else { child = get_dentry_child_with_utf16le_name(dentry, utf16le_name, - utf16le_name_nbytes); + utf16le_name_nbytes, + case_type); FREE(utf16le_name); } return child; @@ -486,46 +799,111 @@ get_dentry_child_with_name(const struct wim_dentry *dentry, const tchar *name) } static struct wim_dentry * -get_dentry_utf16le(WIMStruct *w, const utf16lechar *path, - size_t path_nbytes) +get_dentry_utf16le(WIMStruct *wim, const utf16lechar *path, + CASE_SENSITIVITY_TYPE case_type) { - struct wim_dentry *cur_dentry, *parent_dentry; - const utf16lechar *p, *pp; + struct wim_dentry *cur_dentry; + const utf16lechar *name_start, *name_end; - cur_dentry = parent_dentry = wim_root_dentry(w); - p = path; - while (1) { - while (*p == cpu_to_le16('/')) - p++; - if (*p == '\0') - break; - pp = p; - while (*pp != cpu_to_le16('/') && *pp != cpu_to_le16('\0')) - pp++; + /* 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 = get_dentry_child_with_utf16le_name(parent_dentry, p, - (void*)pp - (void*)p); - if (cur_dentry == NULL) - break; - p = pp; - parent_dentry = cur_dentry; - } - if (cur_dentry == NULL) { - if (dentry_is_directory(parent_dentry)) + name_start = path; + for (;;) { + if (cur_dentry == NULL) { errno = ENOENT; - else + return NULL; + } + + if (*name_start && !dentry_is_directory(cur_dentry)) { errno = ENOTDIR; + return NULL; + } + + 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 cur_dentry; } -/* Returns the dentry corresponding to the @path, or NULL if there is no such - * 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 *w, const tchar *path) +get_dentry(WIMStruct *wim, const tchar *path, CASE_SENSITIVITY_TYPE case_type) { #if TCHAR_IS_UTF16LE - return get_dentry_utf16le(w, path, tstrlen(path) * sizeof(tchar)); + return get_dentry_utf16le(wim, path, case_type); #else utf16lechar *path_utf16le; size_t path_utf16le_nbytes; @@ -536,49 +914,41 @@ get_dentry(WIMStruct *w, const tchar *path) &path_utf16le, &path_utf16le_nbytes); if (ret) return NULL; - dentry = get_dentry_utf16le(w, path_utf16le, path_utf16le_nbytes); + dentry = get_dentry_utf16le(wim, path_utf16le, case_type); FREE(path_utf16le); return dentry; #endif } -struct wim_inode * -wim_pathname_to_inode(WIMStruct *w, const tchar *path) -{ - struct wim_dentry *dentry; - dentry = get_dentry(w, path); - if (dentry) - return dentry->d_inode; - else - return NULL; -} - /* Takes in a path of length @len in @buf, and transforms it into a string for * the path of its parent directory. */ static void to_parent_name(tchar *buf, size_t len) { ssize_t i = (ssize_t)len - 1; - while (i >= 0 && buf[i] == T('/')) + while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR) i--; - while (i >= 0 && buf[i] != T('/')) + while (i >= 0 && buf[i] != WIM_PATH_SEPARATOR) i--; - while (i >= 0 && buf[i] == T('/')) + while (i >= 0 && buf[i] == WIM_PATH_SEPARATOR) i--; buf[i + 1] = T('\0'); } -/* Returns the dentry that corresponds to the parent directory of @path, or NULL - * if the dentry is not found. */ +/* 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 *w, const tchar *path) +get_parent_dentry(WIMStruct *wim, const tchar *path, + CASE_SENSITIVITY_TYPE case_type) { 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(w, buf); + return get_dentry(wim, buf, case_type); } /* Prints the full path of a dentry. */ @@ -703,7 +1073,7 @@ dentry_common_init(struct wim_dentry *dentry) } struct wim_inode * -new_timeless_inode() +new_timeless_inode(void) { struct wim_inode *inode = CALLOC(1, sizeof(struct wim_inode)); if (inode) { @@ -711,20 +1081,15 @@ new_timeless_inode() inode->i_nlink = 1; inode->i_next_stream_id = 1; inode->i_not_rpfixed = 1; - #ifdef WITH_FUSE - if (pthread_mutex_init(&inode->i_mutex, NULL) != 0) { - ERROR_WITH_ERRNO("Error initializing mutex"); - FREE(inode); - return NULL; - } - #endif + inode->i_canonical_streams = 1; + INIT_LIST_HEAD(&inode->i_list); INIT_LIST_HEAD(&inode->i_dentry); } return inode; } static struct wim_inode * -new_inode() +new_inode(void) { struct wim_inode *inode = new_timeless_inode(); if (inode) { @@ -744,7 +1109,7 @@ new_dentry(const tchar *name, struct wim_dentry **dentry_ret) int ret; dentry = MALLOC(sizeof(struct wim_dentry)); - if (!dentry) + if (dentry == NULL) return WIMLIB_ERR_NOMEM; dentry_common_init(dentry); @@ -762,7 +1127,7 @@ new_dentry(const tchar *name, struct wim_dentry **dentry_ret) static int -__new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret, +_new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret, bool timeless) { struct wim_dentry *dentry; @@ -776,7 +1141,7 @@ __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret, dentry->d_inode = new_timeless_inode(); else dentry->d_inode = new_inode(); - if (!dentry->d_inode) { + if (dentry->d_inode == NULL) { free_dentry(dentry); return WIMLIB_ERR_NOMEM; } @@ -789,15 +1154,45 @@ __new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret, int new_dentry_with_timeless_inode(const tchar *name, struct wim_dentry **dentry_ret) { - return __new_dentry_with_inode(name, dentry_ret, true); + return _new_dentry_with_inode(name, dentry_ret, true); } int new_dentry_with_inode(const tchar *name, struct wim_dentry **dentry_ret) { - return __new_dentry_with_inode(name, dentry_ret, false); + return _new_dentry_with_inode(name, dentry_ret, false); +} + +int +new_filler_directory(const tchar *name, struct wim_dentry **dentry_ret) +{ + int ret; + struct wim_dentry *dentry; + + DEBUG("Creating filler directory \"%"TS"\"", name); + ret = new_dentry_with_inode(name, &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; +} + +static int +dentry_clear_inode_visited(struct wim_dentry *dentry, void *_ignore) +{ + dentry->d_inode->i_visited = 0; + return 0; } +void +dentry_tree_clear_inode_visited(struct wim_dentry *root) +{ + for_dentry_in_tree(root, dentry_clear_inode_visited, NULL); +} static int init_ads_entry(struct wim_ads_entry *ads_entry, const void *name, @@ -808,10 +1203,10 @@ init_ads_entry(struct wim_ads_entry *ads_entry, const void *name, if (is_utf16le) { utf16lechar *p = MALLOC(name_nbytes + sizeof(utf16lechar)); - if (!p) + if (p == NULL) return WIMLIB_ERR_NOMEM; memcpy(p, name, name_nbytes); - p[name_nbytes / 2] = 0; + p[name_nbytes / 2] = cpu_to_le16(0); ads_entry->stream_name = p; ads_entry->stream_name_nbytes = name_nbytes; } else { @@ -839,14 +1234,10 @@ free_inode(struct wim_inode *inode) destroy_ads_entry(&inode->i_ads_entries[i]); FREE(inode->i_ads_entries); } - #ifdef WITH_FUSE - wimlib_assert(inode->i_num_opened_fds == 0); - FREE(inode->i_fds); - pthread_mutex_destroy(&inode->i_mutex); - if (inode->i_hlist.pprev) + /* HACK: This may instead delete the inode from i_list, but the + * hlist_del() behaves the same as list_del(). */ + if (!hlist_unhashed(&inode->i_hlist)) hlist_del(&inode->i_hlist); - #endif - FREE(inode->i_extracted_file); FREE(inode); } } @@ -875,27 +1266,28 @@ put_inode(struct wim_inode *inode) void free_dentry(struct wim_dentry *dentry) { - FREE(dentry->file_name); - FREE(dentry->short_name); - FREE(dentry->_full_path); - if (dentry->d_inode) - put_inode(dentry->d_inode); - FREE(dentry); + if (dentry) { + FREE(dentry->file_name); + FREE(dentry->short_name); + FREE(dentry->_full_path); + if (dentry->d_inode) + put_inode(dentry->d_inode); + 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 *_lookup_table) { - struct wim_lookup_table *lookup_table = __lookup_table; - unsigned i; + struct wim_lookup_table *lookup_table = _lookup_table; if (lookup_table) { - struct wim_lookup_table_entry *lte; struct wim_inode *inode = dentry->d_inode; - wimlib_assert(inode->i_nlink != 0); - for (i = 0; i <= inode->i_num_ads; i++) { + for (unsigned i = 0; i <= inode->i_num_ads; i++) { + struct wim_lookup_table_entry *lte; + lte = inode_stream_lte(inode, i, lookup_table); if (lte) lte_decrement_refcnt(lte, lookup_table); @@ -908,38 +1300,84 @@ do_free_dentry(struct wim_dentry *dentry, void *__lookup_table) /* * 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. + * @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 wim_dentry *root, struct wim_lookup_table *lookup_table) { - if (root) - for_dentry_in_tree_depth(root, do_free_dentry, lookup_table); + for_dentry_in_tree_depth(root, do_free_dentry, lookup_table); +} + +/* Insert a dentry into the case insensitive index for a directory. + * + * This is a red-black tree, but when multiple dentries share the same + * case-insensitive name, only one is inserted into the tree itself; the rest + * are connected in a list. + */ +static struct wim_dentry * +dentry_add_child_case_insensitive(struct wim_dentry *parent, + struct wim_dentry *child) +{ + struct rb_root *root; + struct rb_node **new; + struct rb_node *rb_parent; + + root = &parent->d_inode->i_children_case_insensitive; + new = &root->rb_node; + rb_parent = NULL; + while (*new) { + struct wim_dentry *this = container_of(*new, struct wim_dentry, + rb_node_case_insensitive); + int result = dentry_compare_names_case_insensitive(child, this); + + rb_parent = *new; + + if (result < 0) + new = &((*new)->rb_left); + else if (result > 0) + new = &((*new)->rb_right); + else + return this; + } + rb_link_node(&child->rb_node_case_insensitive, rb_parent, new); + rb_insert_color(&child->rb_node_case_insensitive, root); + return NULL; } /* * Links a dentry into the directory tree. * - * @parent: The dentry that will be the parent of @dentry. - * @dentry: The dentry to link. + * @parent: The dentry that will be the parent of @child. + * @child: The dentry to link. + * + * Returns NULL if successful. If @parent already contains a dentry with the + * same case-sensitive name as @child, the pointer to this duplicate dentry is + * returned. */ struct wim_dentry * dentry_add_child(struct wim_dentry * restrict parent, struct wim_dentry * restrict child) { - wimlib_assert(dentry_is_directory(parent)); + struct rb_root *root; + struct rb_node **new; + struct rb_node *rb_parent; - struct rb_root *root = &parent->d_inode->i_children; - struct rb_node **new = &(root->rb_node); - struct rb_node *rb_parent = NULL; + wimlib_assert(dentry_is_directory(parent)); + wimlib_assert(parent != child); + /* Case sensitive child dentry index */ + root = &parent->d_inode->i_children; + new = &root->rb_node; + rb_parent = NULL; while (*new) { struct wim_dentry *this = rbnode_dentry(*new); - int result = dentry_compare_names(child, this); + int result = dentry_compare_names_case_sensitive(child, this); rb_parent = *new; @@ -953,6 +1391,19 @@ dentry_add_child(struct wim_dentry * restrict parent, child->parent = parent; rb_link_node(&child->rb_node, rb_parent, new); rb_insert_color(&child->rb_node, root); + + /* Case insensitive child dentry index */ + { + struct wim_dentry *existing; + existing = dentry_add_child_case_insensitive(parent, child); + if (existing) { + list_add(&child->case_insensitive_conflict_list, + &existing->case_insensitive_conflict_list); + child->rb_node_case_insensitive.__rb_parent_color = 0; + } else { + INIT_LIST_HEAD(&child->case_insensitive_conflict_list); + } + } return NULL; } @@ -961,14 +1412,108 @@ void unlink_dentry(struct wim_dentry *dentry) { struct wim_dentry *parent = dentry->parent; + if (parent == dentry) return; rb_erase(&dentry->rb_node, &parent->d_inode->i_children); + + if (dentry->rb_node_case_insensitive.__rb_parent_color) { + /* This dentry was in the case-insensitive red-black tree. */ + rb_erase(&dentry->rb_node_case_insensitive, + &parent->d_inode->i_children_case_insensitive); + if (!list_empty(&dentry->case_insensitive_conflict_list)) { + /* Make a different case-insensitively-the-same dentry + * be the "representative" in the red-black tree. */ + struct list_head *next; + struct wim_dentry *other; + struct wim_dentry *existing; + + next = dentry->case_insensitive_conflict_list.next; + other = list_entry(next, struct wim_dentry, case_insensitive_conflict_list); + existing = dentry_add_child_case_insensitive(parent, other); + wimlib_assert(existing == NULL); + } + } + list_del(&dentry->case_insensitive_conflict_list); +} + +static int +free_dentry_full_path(struct wim_dentry *dentry, void *_ignore) +{ + FREE(dentry->_full_path); + dentry->_full_path = NULL; + return 0; +} + +/* Rename a file or directory in the WIM. */ +int +rename_wim_path(WIMStruct *wim, const tchar *from, const tchar *to, + CASE_SENSITIVITY_TYPE case_type) +{ + struct wim_dentry *src; + struct wim_dentry *dst; + struct wim_dentry *parent_of_dst; + int ret; + + /* This rename() implementation currently only supports actual files + * (not alternate data streams) */ + + src = get_dentry(wim, from, case_type); + if (!src) + return -errno; + + dst = get_dentry(wim, to, case_type); + + if (dst) { + /* Destination file exists */ + + if (src == dst) /* Same file */ + return 0; + + if (!dentry_is_directory(src)) { + /* Cannot rename non-directory to directory. */ + if (dentry_is_directory(dst)) + return -EISDIR; + } else { + /* Cannot rename directory to a non-directory or a non-empty + * directory */ + if (!dentry_is_directory(dst)) + return -ENOTDIR; + if (dentry_has_children(dst)) + return -ENOTEMPTY; + } + parent_of_dst = dst->parent; + } else { + /* Destination does not exist */ + parent_of_dst = get_parent_dentry(wim, to, case_type); + if (!parent_of_dst) + return -errno; + + if (!dentry_is_directory(parent_of_dst)) + return -ENOTDIR; + } + + ret = set_dentry_name(src, path_basename(to)); + if (ret) + return -ENOMEM; + if (dst) { + unlink_dentry(dst); + free_dentry_tree(dst, wim->lookup_table); + } + unlink_dentry(src); + dentry_add_child(parent_of_dst, src); + if (src->_full_path) + for_dentry_in_tree(src, free_dentry_full_path, NULL); + return 0; } /* * Returns the alternate data stream entry belonging to @inode that has the - * stream name @stream_name. + * stream name @stream_name, or NULL if the inode has no alternate data stream + * with that name. + * + * If @p stream_name is the empty string, NULL is returned --- that is, this + * function will not return "unnamed" alternate data stream entries. */ struct wim_ads_entry * inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name, @@ -981,6 +1526,9 @@ inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name, u16 i; struct wim_ads_entry *result; + if (stream_name[0] == T('\0')) + return NULL; + #if TCHAR_IS_UTF16LE const utf16lechar *stream_name_utf16le; @@ -1004,7 +1552,8 @@ inode_get_ads_entry(struct wim_inode *inode, const tchar *stream_name, do { if (ads_entry_has_name(&inode->i_ads_entries[i], stream_name_utf16le, - stream_name_utf16le_nbytes)) + stream_name_utf16le_nbytes, + default_ignore_case)) { if (idx_ret) *idx_ret = i; @@ -1027,6 +1576,8 @@ do_inode_add_ads(struct wim_inode *inode, const void *stream_name, struct wim_ads_entry *ads_entries; struct wim_ads_entry *new_entry; + wimlib_assert(stream_name_nbytes != 0); + if (inode->i_num_ads >= 0xfffe) { ERROR("Too many alternate data streams in one inode!"); return NULL; @@ -1034,7 +1585,7 @@ do_inode_add_ads(struct wim_inode *inode, const void *stream_name, num_ads = inode->i_num_ads + 1; ads_entries = REALLOC(inode->i_ads_entries, num_ads * sizeof(inode->i_ads_entries[0])); - if (!ads_entries) { + if (ads_entries == NULL) { ERROR("Failed to allocate memory for new alternate data stream"); return NULL; } @@ -1058,8 +1609,10 @@ inode_add_ads_utf16le(struct wim_inode *inode, } /* - * Add an alternate stream entry to a WIM inode and return a pointer to it, or - * NULL if memory could not be allocated. + * Add an alternate stream entry to a WIM inode. On success, returns a pointer + * to the new entry; on failure, returns NULL. + * + * @stream_name must be a nonempty string. */ struct wim_ads_entry * inode_add_ads(struct wim_inode *inode, const tchar *stream_name) @@ -1070,67 +1623,96 @@ inode_add_ads(struct wim_inode *inode, const tchar *stream_name) TCHAR_IS_UTF16LE); } -int -inode_add_ads_with_data(struct wim_inode *inode, const tchar *name, - const void *value, size_t size, - struct wim_lookup_table *lookup_table) +static struct wim_lookup_table_entry * +add_stream_from_data_buffer(const void *buffer, size_t size, + struct wim_lookup_table *lookup_table) { - int ret = WIMLIB_ERR_NOMEM; - struct wim_ads_entry *new_ads_entry; - struct wim_lookup_table_entry *existing_lte; - struct wim_lookup_table_entry *lte; - u8 value_hash[SHA1_HASH_SIZE]; + u8 hash[SHA1_HASH_SIZE]; + struct wim_lookup_table_entry *lte, *existing_lte; - wimlib_assert(inode->i_resolved); - new_ads_entry = inode_add_ads(inode, name); - if (!new_ads_entry) - goto out; - sha1_buffer((const u8*)value, size, value_hash); - existing_lte = __lookup_resource(lookup_table, value_hash); + sha1_buffer(buffer, size, hash); + existing_lte = lookup_resource(lookup_table, hash); if (existing_lte) { + wimlib_assert(existing_lte->size == size); lte = existing_lte; lte->refcnt++; } else { - u8 *value_copy; + void *buffer_copy; lte = new_lookup_table_entry(); - if (!lte) - goto out_remove_ads_entry; - value_copy = MALLOC(size); - if (!value_copy) { - FREE(lte); - goto out_remove_ads_entry; + if (lte == NULL) + return NULL; + buffer_copy = memdup(buffer, size); + if (buffer_copy == NULL) { + free_lookup_table_entry(lte); + return NULL; } - memcpy(value_copy, value, size); - lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; - lte->attached_buffer = value_copy; - lte->resource_entry.original_size = size; - lte->resource_entry.size = size; - copy_hash(lte->hash, value_hash); + lte->resource_location = RESOURCE_IN_ATTACHED_BUFFER; + lte->attached_buffer = buffer_copy; + lte->size = size; + copy_hash(lte->hash, hash); lookup_table_insert(lookup_table, lte); } - new_ads_entry->lte = lte; - ret = 0; - goto out; -out_remove_ads_entry: - inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries, - lookup_table); -out: - return ret; + return lte; } -/* Remove an alternate data stream from a WIM inode */ -void -inode_remove_ads(struct wim_inode *inode, u16 idx, - struct wim_lookup_table *lookup_table) +int +inode_add_ads_with_data(struct wim_inode *inode, const tchar *name, + const void *value, size_t size, + struct wim_lookup_table *lookup_table) { - struct wim_ads_entry *ads_entry; - struct wim_lookup_table_entry *lte; + struct wim_ads_entry *new_ads_entry; - wimlib_assert(idx < inode->i_num_ads); wimlib_assert(inode->i_resolved); - ads_entry = &inode->i_ads_entries[idx]; - + new_ads_entry = inode_add_ads(inode, name); + if (new_ads_entry == NULL) + return WIMLIB_ERR_NOMEM; + + new_ads_entry->lte = add_stream_from_data_buffer(value, size, + lookup_table); + if (new_ads_entry->lte == NULL) { + inode_remove_ads(inode, new_ads_entry - inode->i_ads_entries, + lookup_table); + return WIMLIB_ERR_NOMEM; + } + return 0; +} + +bool +inode_has_named_stream(const struct wim_inode *inode) +{ + for (u16 i = 0; i < inode->i_num_ads; i++) + if (ads_entry_is_named_stream(&inode->i_ads_entries[i])) + return true; + return false; +} + +/* Set the unnamed stream of a WIM inode, given a data buffer containing the + * stream contents. */ +int +inode_set_unnamed_stream(struct wim_inode *inode, const void *data, size_t len, + struct wim_lookup_table *lookup_table) +{ + inode->i_lte = add_stream_from_data_buffer(data, len, lookup_table); + if (inode->i_lte == NULL) + return WIMLIB_ERR_NOMEM; + inode->i_resolved = 1; + return 0; +} + +/* Remove an alternate data stream from a WIM inode */ +void +inode_remove_ads(struct wim_inode *inode, u16 idx, + struct wim_lookup_table *lookup_table) +{ + struct wim_ads_entry *ads_entry; + struct wim_lookup_table_entry *lte; + + wimlib_assert(idx < inode->i_num_ads); + wimlib_assert(inode->i_resolved); + + ads_entry = &inode->i_ads_entries[idx]; + DEBUG("Remove alternate data stream \"%"WS"\"", ads_entry->stream_name); lte = ads_entry->lte; @@ -1145,6 +1727,15 @@ inode_remove_ads(struct wim_inode *inode, u16 idx, inode->i_num_ads--; } +bool +inode_has_unix_data(const struct wim_inode *inode) +{ + for (u16 i = 0; i < inode->i_num_ads; i++) + if (ads_entry_is_unix_data(&inode->i_ads_entries[i])) + return true; + return false; +} + #ifndef __WIN32__ int inode_get_unix_data(const struct wim_inode *inode, @@ -1160,21 +1751,21 @@ inode_get_unix_data(const struct wim_inode *inode, ads_entry = inode_get_ads_entry((struct wim_inode*)inode, WIMLIB_UNIX_DATA_TAG, NULL); - if (!ads_entry) + if (ads_entry == NULL) return NO_UNIX_DATA; if (stream_idx_ret) *stream_idx_ret = ads_entry - inode->i_ads_entries; lte = ads_entry->lte; - if (!lte) + if (lte == NULL) return NO_UNIX_DATA; - size = wim_resource_size(lte); + size = lte->size; if (size != sizeof(struct wimlib_unix_data)) return BAD_UNIX_DATA; - ret = read_full_resource_into_buf(lte, unix_data, true); + ret = read_full_stream_into_buf(lte, unix_data); if (ret) return ret; @@ -1217,182 +1808,141 @@ inode_set_unix_data(struct wim_inode *inode, uid_t uid, gid_t gid, mode_t mode, } #endif /* !__WIN32__ */ -static void -replace_forbidden_characters(utf16lechar *name) -{ - utf16lechar *p; - - for (p = name; *p; p++) { - #ifdef __WIN32__ - if (wcschr(L"<>:\"/\\|?*", (wchar_t)*p)) - #else - if (*p == '/') - #endif - { - if (name) { - WARNING("File, directory, or stream name \"%"WS"\"\n" - " contains forbidden characters; " - "replacing them with Unicode codepoint U+001A", - name); - name = NULL; - } - *p = 0x1a; - } - } -} - /* * Reads the alternate data stream entries of a WIM dentry. * - * @p: Pointer to buffer that starts with the first alternate stream entry. + * @p: + * Pointer to buffer that starts with the first alternate stream entry. * - * @inode: Inode to load the alternate data streams into. - * @inode->i_num_ads must have been set to the number of - * alternate data streams that are expected. + * @inode: + * Inode to load the alternate data streams into. @inode->i_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. + * @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: + * On success, inode->i_ads_entries is set to an array of `struct + * wim_ads_entry's of length inode->i_num_ads. On failure, @inode is not + * modified. * - * struct wim_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->i_ads_entries - * is set to an array of `struct wim_ads_entry's of length inode->i_num_ads. On - * failure, @inode is not modified. + * Return values: + * WIMLIB_ERR_SUCCESS (0) + * WIMLIB_ERR_INVALID_METADATA_RESOURCE + * WIMLIB_ERR_NOMEM */ static int -read_ads_entries(const u8 *p, struct wim_inode *inode, u64 remaining_size) +read_ads_entries(const u8 * restrict p, struct wim_inode * restrict inode, + size_t nbytes_remaining) { u16 num_ads; struct wim_ads_entry *ads_entries; int ret; + BUILD_BUG_ON(sizeof(struct wim_ads_entry_on_disk) != WIM_ADS_ENTRY_DISK_SIZE); + + /* Allocate an array for our in-memory representation of the alternate + * data stream entries. */ num_ads = inode->i_num_ads; ads_entries = CALLOC(num_ads, sizeof(inode->i_ads_entries[0])); - if (!ads_entries) { - ERROR("Could not allocate memory for %"PRIu16" " - "alternate data stream entries", num_ads); - return WIMLIB_ERR_NOMEM; - } + if (ads_entries == NULL) + goto out_of_memory; + /* Read the entries into our newly allocated buffer. */ for (u16 i = 0; i < num_ads; i++) { - struct wim_ads_entry *cur_entry; u64 length; - u64 length_no_padding; - u64 total_length; - const u8 *p_save = p; + struct wim_ads_entry *cur_entry; + const struct wim_ads_entry_on_disk *disk_entry = + (const struct wim_ads_entry_on_disk*)p; cur_entry = &ads_entries[i]; - - #ifdef WITH_FUSE ads_entries[i].stream_id = i + 1; - #endif - - /* 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; - } - - p = get_u64(p, &length); - p += 8; /* Skip the reserved field */ - p = get_bytes(p, SHA1_HASH_SIZE, cur_entry->hash); - p = get_u16(p, &cur_entry->stream_name_nbytes); - - cur_entry->stream_name = NULL; - - /* Length including neither the null terminator nor the padding - * */ - length_no_padding = WIM_ADS_ENTRY_DISK_SIZE + - cur_entry->stream_name_nbytes; - /* Length including the null terminator and the padding */ - total_length = ((length_no_padding + 2) + 7) & ~7; - - wimlib_assert(total_length == ads_entry_total_length(cur_entry)); + /* Do we have at least the size of the fixed-length data we know + * need? */ + if (nbytes_remaining < sizeof(struct wim_ads_entry_on_disk)) + goto out_invalid; + + /* Read the length field */ + length = le64_to_cpu(disk_entry->length); + + /* 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_ads_entry_on_disk) || + length > nbytes_remaining) + goto out_invalid; + + /* Read the rest of the fixed-length data. */ + + cur_entry->reserved = le64_to_cpu(disk_entry->reserved); + copy_hash(cur_entry->hash, disk_entry->hash); + cur_entry->stream_name_nbytes = le16_to_cpu(disk_entry->stream_name_nbytes); + + /* If stream_name_nbytes != 0, this is a named stream. + * Otherwise this is an unnamed stream, or in some cases (bugs + * in Microsoft's software I guess) a meaningless entry + * distinguished from the real unnamed stream entry, if any, by + * the fact that the real unnamed stream entry has a nonzero + * hash field. */ + if (cur_entry->stream_name_nbytes) { + /* 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 (cur_entry->stream_name_nbytes & 1) + goto out_invalid; + + /* 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_ads_entry_on_disk) + + cur_entry->stream_name_nbytes > length) + goto out_invalid; - 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; - } + cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2); + if (cur_entry->stream_name == NULL) + goto out_of_memory; - /* 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; + memcpy(cur_entry->stream_name, + disk_entry->stream_name, + cur_entry->stream_name_nbytes); + cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = cpu_to_le16(0); + } else { + /* Mark inode as having weird stream entries. */ + inode->i_canonical_streams = 0; } - if (cur_entry->stream_name_nbytes) { - cur_entry->stream_name = MALLOC(cur_entry->stream_name_nbytes + 2); - if (!cur_entry->stream_name) { - ret = WIMLIB_ERR_NOMEM; - goto out_free_ads_entries; - } - get_bytes(p, cur_entry->stream_name_nbytes, - cur_entry->stream_name); - cur_entry->stream_name[cur_entry->stream_name_nbytes / 2] = 0; - replace_forbidden_characters(cur_entry->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; + * pointer by @length to reach the next ADS entry, it's possible + * that less than @length is actually remaining in the metadata + * resource. We should set the remaining bytes to 0 if this + * happens. */ + length = (length + 7) & ~(u64)7; + p += length; + if (nbytes_remaining < length) + nbytes_remaining = 0; else - remaining_size -= total_length; + nbytes_remaining -= length; } inode->i_ads_entries = ads_entries; -#ifdef WITH_FUSE inode->i_next_stream_id = inode->i_num_ads + 1; -#endif - return 0; + ret = 0; + goto out; +out_of_memory: + ret = WIMLIB_ERR_NOMEM; + goto out_free_ads_entries; +out_invalid: + ERROR("An alternate data stream entry is invalid"); + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; out_free_ads_entries: - for (u16 i = 0; i < num_ads; i++) - destroy_ads_entry(&ads_entries[i]); - FREE(ads_entries); + if (ads_entries) { + for (u16 i = 0; i < num_ads; i++) + destroy_ads_entry(&ads_entries[i]); + FREE(ads_entries); + } +out: return ret; } @@ -1400,9 +1950,14 @@ out_free_ads_entries: * Reads a WIM 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. + * @metadata_resource: + * Pointer to the metadata resource buffer. + * + * @metadata_resource_len: + * Length of the metadata resource buffer, in bytes. + * + * @offset: Offset of the dentry within the metadata resource. + * * @dentry: A `struct wim_dentry' that will be filled in by this function. * * Return 0 on success or nonzero on failure. On failure, @dentry will have @@ -1410,102 +1965,123 @@ out_free_ads_entries: * 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. + * + * Return values: + * WIMLIB_ERR_SUCCESS (0) + * WIMLIB_ERR_INVALID_METADATA_RESOURCE + * WIMLIB_ERR_NOMEM */ int -read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, - u64 offset, struct wim_dentry *dentry) +read_dentry(const u8 * restrict metadata_resource, u64 metadata_resource_len, + u64 offset, struct wim_dentry * restrict dentry) { - const u8 *p; + u64 calculated_size; - utf16lechar *file_name = NULL; - utf16lechar *short_name = NULL; + utf16lechar *file_name; + utf16lechar *short_name; u16 short_name_nbytes; u16 file_name_nbytes; int ret; - struct wim_inode *inode = NULL; + struct wim_inode *inode; + const u8 *p = &metadata_resource[offset]; + const struct wim_dentry_on_disk *disk_dentry = + (const struct wim_dentry_on_disk*)p; + + BUILD_BUG_ON(sizeof(struct wim_dentry_on_disk) != WIM_DENTRY_DISK_SIZE); + + if ((uintptr_t)p & 7) + WARNING("WIM dentry is not 8-byte aligned"); dentry_common_init(dentry); - /*Make sure the dentry really fits into the metadata resource.*/ - if (offset + 8 > metadata_resource_len || offset + 8 < offset) { + /* 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. */ + if (offset + sizeof(u64) > metadata_resource_len || + offset + sizeof(u64) < 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; + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; } - - /* 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); + dentry->length = le64_to_cpu(disk_dentry->length); /* 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. */ + * return successfully in this case. */ + if (dentry->length == 8) + dentry->length = 0; if (dentry->length == 0) return 0; - /* 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) + /* Now that we have the actual length provided in the on-disk structure, + * again make sure it doesn't overflow the metadata resource buffer. */ + 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; + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; } - if (dentry->length < WIM_DENTRY_DISK_SIZE) { + /* Make sure the dentry length is at least as large as the number of + * fixed-length fields */ + if (dentry->length < sizeof(struct wim_dentry_on_disk)) { ERROR("Directory entry has invalid length of %"PRIu64" bytes", dentry->length); - return WIMLIB_ERR_INVALID_DENTRY; + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; } + /* Allocate a `struct wim_inode' for this `struct wim_dentry'. */ inode = new_timeless_inode(); - if (!inode) + if (inode == NULL) return WIMLIB_ERR_NOMEM; - p = get_u32(p, &inode->i_attributes); - p = get_u32(p, (u32*)&inode->i_security_id); - p = get_u64(p, &dentry->subdir_offset); - - /* 2 unused fields */ - p += 2 * sizeof(u64); - /*p = get_u64(p, &dentry->unused1);*/ - /*p = get_u64(p, &dentry->unused2);*/ - - p = get_u64(p, &inode->i_creation_time); - p = get_u64(p, &inode->i_last_access_time); - p = get_u64(p, &inode->i_last_write_time); + /* Read more fields; some into the dentry, and some into the inode. */ - p = get_bytes(p, SHA1_HASH_SIZE, inode->i_hash); + 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_unused_1 = le64_to_cpu(disk_dentry->unused_1); + dentry->d_unused_2 = le64_to_cpu(disk_dentry->unused_2); + 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); /* 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. */ + * document it. So we have some fields we read for reparse points, and + * some fields in the same place for non-reparse-point.s */ if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) { - p = get_u32(p, &inode->i_rp_unknown_1); - p = get_u32(p, &inode->i_reparse_tag); - p = get_u16(p, &inode->i_rp_unknown_2); - p = get_u16(p, &inode->i_not_rpfixed); + 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. */ } else { - p = get_u32(p, &inode->i_rp_unknown_1); - p = get_u64(p, &inode->i_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) */ + inode->i_num_ads = le16_to_cpu(disk_dentry->num_alternate_data_streams); - p = get_u16(p, &inode->i_num_ads); + short_name_nbytes = le16_to_cpu(disk_dentry->short_name_nbytes); + file_name_nbytes = le16_to_cpu(disk_dentry->file_name_nbytes); - p = get_u16(p, &short_name_nbytes); - p = get_u16(p, &file_name_nbytes); + if ((short_name_nbytes & 1) | (file_name_nbytes & 1)) + { + ERROR("Dentry name is not valid UTF-16LE (odd number of bytes)!"); + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; + goto out_free_inode; + } /* 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. @@ -1513,128 +2089,78 @@ read_dentry(const u8 metadata_resource[], u64 metadata_resource_len, * 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); + calculated_size = dentry_correct_length_unaligned(file_name_nbytes, + short_name_nbytes); if (dentry->length < calculated_size) { ERROR("Unexpected end of directory entry! (Expected " - "at least %"PRIu64" bytes, got %"PRIu64" bytes. " - "short_name_nbytes = %hu, file_name_nbytes = %hu)", - calculated_size, dentry->length, - short_name_nbytes, file_name_nbytes); - ret = WIMLIB_ERR_INVALID_DENTRY; + "at least %"PRIu64" bytes, got %"PRIu64" bytes.)", + calculated_size, dentry->length); + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; goto out_free_inode; } + p += sizeof(struct wim_dentry_on_disk); + /* Read the filename if present. Note: if the filename is empty, there * is no null terminator following it. */ if (file_name_nbytes) { file_name = MALLOC(file_name_nbytes + 2); - if (!file_name) { + if (file_name == NULL) { ERROR("Failed to allocate %d bytes for dentry file name", file_name_nbytes + 2); ret = WIMLIB_ERR_NOMEM; goto out_free_inode; } - p = get_bytes(p, file_name_nbytes + 2, file_name); - if (file_name[file_name_nbytes / 2] != 0) { - file_name[file_name_nbytes / 2] = 0; - WARNING("File name in WIM dentry \"%"WS"\" is not " - "null-terminated!", file_name); - } - replace_forbidden_characters(file_name); + memcpy(file_name, p, file_name_nbytes); + p += file_name_nbytes + 2; + file_name[file_name_nbytes / 2] = cpu_to_le16(0); + } else { + file_name = NULL; } - /* Align the calculated size */ - calculated_size = (calculated_size + 7) & ~7; - - 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 `%"WS"' has %"PRIu64" "*/ - /*"extra bytes of data", file_name,*/ - /*dentry->length - calculated_size);*/ - } /* Read the short filename if present. Note: if there is no short * filename, there is no null terminator following it. */ if (short_name_nbytes) { short_name = MALLOC(short_name_nbytes + 2); - if (!short_name) { + if (short_name == NULL) { ERROR("Failed to allocate %d bytes for dentry short name", short_name_nbytes + 2); ret = WIMLIB_ERR_NOMEM; goto out_free_file_name; } - p = get_bytes(p, short_name_nbytes + 2, short_name); - if (short_name[short_name_nbytes / 2] != 0) { - short_name[short_name_nbytes / 2] = 0; - WARNING("Short name in WIM dentry \"%"WS"\" is not " - "null-terminated!", file_name); - } - replace_forbidden_characters(short_name); + memcpy(short_name, p, short_name_nbytes); + p += short_name_nbytes + 2; + short_name[short_name_nbytes / 2] = cpu_to_le16(0); + } else { + short_name = NULL; } + /* Align the dentry length */ + dentry->length = (dentry->length + 7) & ~7; + /* * 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. + * aligned boundary, and the alternate data stream entries seem to NOT + * be included in the dentry->length field for some reason. */ if (inode->i_num_ads != 0) { - - /* Trying different lengths is just a hack to make sure we have - * a chance of reading the ADS entries correctly despite the - * poor documentation. */ - - if (calculated_size != dentry->length) { - WARNING("Trying calculated dentry length (%"PRIu64") " - "instead of dentry->length field (%"PRIu64") " - "to read ADS entries", - calculated_size, dentry->length); - } - u64 lengths_to_try[3] = {calculated_size, - (dentry->length + 7) & ~7, - dentry->length}; - ret = WIMLIB_ERR_INVALID_DENTRY; - for (size_t i = 0; i < ARRAY_LEN(lengths_to_try); i++) { - if (lengths_to_try[i] > metadata_resource_len - offset) - continue; - ret = read_ads_entries(&metadata_resource[offset + lengths_to_try[i]], - inode, - metadata_resource_len - offset - lengths_to_try[i]); - if (ret == 0) - goto out; + ret = WIMLIB_ERR_INVALID_METADATA_RESOURCE; + if (offset + dentry->length > metadata_resource_len || + (ret = read_ads_entries(&metadata_resource[offset + dentry->length], + inode, + metadata_resource_len - offset - dentry->length))) + { + ERROR("Failed to read alternate data stream " + "entries of WIM dentry \"%"WS"\"", file_name); + goto out_free_short_name; } - ERROR("Failed to read alternate data stream " - "entries of WIM dentry \"%"WS"\"", file_name); - goto out_free_short_name; } -out: /* We've read all the data for this dentry. Set the names and their * lengths, and we've done. */ dentry->d_inode = inode; @@ -1642,38 +2168,59 @@ out: dentry->short_name = short_name; dentry->file_name_nbytes = file_name_nbytes; dentry->short_name_nbytes = short_name_nbytes; - return 0; + ret = 0; + goto out; out_free_short_name: FREE(short_name); out_free_file_name: FREE(file_name); out_free_inode: free_inode(inode); +out: 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"); +} + /* 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. + * @metadata_resource: + * An array that contains the uncompressed metadata resource for the WIM + * file. * - * @metadata_resource_len: The length of the uncompressed metadata resource, in - * bytes. + * @metadata_resource_len: + * The length of the uncompressed metadata resource, in bytes. * - * @dentry: A pointer to a `struct wim_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. + * @dentry: + * A pointer to a `struct wim_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. * - * Returns zero on success; nonzero on failure. + * Return values: + * WIMLIB_ERR_SUCCESS (0) + * WIMLIB_ERR_INVALID_METADATA_RESOURCE + * WIMLIB_ERR_NOMEM */ int -read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len, - struct wim_dentry *dentry) +read_dentry_tree(const u8 * restrict metadata_resource, + u64 metadata_resource_len, + struct wim_dentry * restrict dentry) { u64 cur_offset = dentry->subdir_offset; struct wim_dentry *child; + struct wim_dentry *duplicate; + struct wim_dentry *parent; struct wim_dentry cur_child; int ret; @@ -1686,13 +2233,25 @@ read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len, if (cur_offset == 0) return 0; + /* Check for cyclic directory structure */ + for (parent = dentry->parent; !dentry_is_root(parent); parent = parent->parent) + { + if (unlikely(parent->subdir_offset == cur_offset)) { + ERROR("Cyclic directory structure detected: children " + "of \"%"TS"\" coincide with children of \"%"TS"\"", + dentry_full_path(dentry), + dentry_full_path(parent)); + return WIMLIB_ERR_INVALID_METADATA_RESOURCE; + } + } + /* Find and read all the children of @dentry. */ - while (1) { + for (;;) { /* Read next child of @dentry into @cur_child. */ ret = read_dentry(metadata_resource, metadata_resource_len, cur_offset, &cur_child); - if (ret != 0) + if (ret) break; /* Check for end of directory. */ @@ -1701,115 +2260,180 @@ read_dentry_tree(const u8 metadata_resource[], u64 metadata_resource_len, /* Not end of directory. Allocate this child permanently and * link it to the parent and previous child. */ - child = MALLOC(sizeof(struct wim_dentry)); - if (!child) { - ERROR("Failed to allocate %zu bytes for new dentry", - sizeof(struct wim_dentry)); + child = memdup(&cur_child, sizeof(struct wim_dentry)); + if (child == NULL) { + ERROR("Failed to allocate new dentry!"); ret = WIMLIB_ERR_NOMEM; break; } - memcpy(child, &cur_child, sizeof(struct wim_dentry)); - dentry_add_child(dentry, child); - inode_add_dentry(child, child->d_inode); - - /* 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; - } /* 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 + * cur_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); + cur_offset += dentry_in_total_length(child); + + if (unlikely(!dentry_has_long_name(child))) { + WARNING("Ignoring unnamed dentry in " + "directory \"%"TS"\"", + dentry_full_path(dentry)); + free_dentry(child); + continue; + } + + duplicate = dentry_add_child(dentry, child); + if (unlikely(duplicate)) { + 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" " + "at that path with the exact same name)", + child_type, dentry_full_path(duplicate), + duplicate_type); + free_dentry(child); + continue; + } + + inode_add_dentry(child, child->d_inode); + /* If there are children of this child, call this + * procedure recursively. */ + if (child->subdir_offset != 0) { + if (likely(dentry_is_directory(child))) { + ret = read_dentry_tree(metadata_resource, + metadata_resource_len, + child); + if (ret) + break; + } else { + WARNING("Ignoring children of non-directory \"%"TS"\"", + dentry_full_path(child)); + } + } } 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) +{ + struct wim_ads_entry_on_disk *disk_ads_entry = + (struct wim_ads_entry_on_disk*)p; + u8 *orig_p = p; + + 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); + } + /* Align to 8-byte boundary */ + while ((uintptr_t)p & 7) + *p++ = 0; + disk_ads_entry->length = cpu_to_le64(p - orig_p); + return p; +} + /* * 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. + * + * Returns the pointer to the byte after the last byte we wrote as part of the + * dentry, including any alternate data stream entries. */ static u8 * -write_dentry(const struct wim_dentry *dentry, u8 *p) +write_dentry(const struct wim_dentry * restrict dentry, u8 * restrict p) { - u8 *orig_p = p; + const struct wim_inode *inode; + struct wim_dentry_on_disk *disk_dentry; + const u8 *orig_p; const u8 *hash; - const struct wim_inode *inode = dentry->d_inode; + bool use_dummy_stream; + u16 num_ads; - /* 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->i_attributes); - p = put_u32(p, inode->i_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->i_creation_time); - p = put_u64(p, inode->i_last_access_time); - p = put_u64(p, inode->i_last_write_time); - hash = inode_stream_hash(inode, 0); - p = put_bytes(p, SHA1_HASH_SIZE, hash); + 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->unused_1 = cpu_to_le64(dentry->d_unused_1); + disk_dentry->unused_2 = cpu_to_le64(dentry->d_unused_2); + 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); if (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT) { - p = put_u32(p, 0); - p = put_u32(p, inode->i_reparse_tag); - p = put_u16(p, 0); - p = put_u16(p, inode->i_not_rpfixed); + 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 { - u64 link_group_id; - p = put_u32(p, 0); - if (inode->i_nlink == 1) - link_group_id = 0; - else - link_group_id = inode->i_ino; - p = put_u64(p, link_group_id); - } - p = put_u16(p, inode->i_num_ads); - p = put_u16(p, dentry->short_name_nbytes); - p = put_u16(p, dentry->file_name_nbytes); - if (dentry_has_long_name(dentry)) { - p = put_bytes(p, dentry->file_name_nbytes + 2, - dentry->file_name); - } - if (dentry_has_short_name(dentry)) { - p = put_bytes(p, dentry->short_name_nbytes + 2, - dentry->short_name); + 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); + 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); + + if (dentry_has_short_name(dentry)) + p = mempcpy(p, dentry->short_name, dentry->short_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)); + 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); - /* 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. */ + if (use_dummy_stream) { + hash = inode_unnamed_stream_hash(inode); + p = write_ads_entry(&(struct wim_ads_entry){}, hash, p); + } + + /* Write the alternate data streams entries, if any. */ for (u16 i = 0; i < inode->i_num_ads; i++) { - p = put_u64(p, ads_entry_total_length(&inode->i_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->i_ads_entries[i].stream_name_nbytes); - if (inode->i_ads_entries[i].stream_name_nbytes) { - p = put_bytes(p, - inode->i_ads_entries[i].stream_name_nbytes + 2, - inode->i_ads_entries[i].stream_name); - } - p = put_zeroes(p, (8 - (p - orig_p) % 8) % 8); + p = write_ads_entry(&inode->i_ads_entries[i], hash, p); } - wimlib_assert(p - orig_p == __dentry_total_length(dentry, length)); + return p; } @@ -1850,7 +2474,8 @@ write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p) for_dentry_child(parent, write_dentry_cb, &p); /* write end of directory entry */ - p = put_u64(p, 0); + *(le64*)p = cpu_to_le64(0); + p += 8; /* Recurse on children. */ for_dentry_child(parent, write_dentry_tree_recursive_cb, &p); @@ -1865,7 +2490,7 @@ write_dentry_tree_recursive(const struct wim_dentry *parent, u8 *p) * Returns pointer to the byte after the last byte we wrote. */ u8 * -write_dentry_tree(const struct wim_dentry *root, u8 *p) +write_dentry_tree(const struct wim_dentry * restrict root, u8 * restrict p) { DEBUG("Writing dentry tree."); wimlib_assert(dentry_is_root(root)); @@ -1876,8 +2501,437 @@ write_dentry_tree(const struct wim_dentry *root, u8 *p) /* 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); + *(le64*)p = cpu_to_le64(0); + p += 8; /* Recursively write the rest of the dentry tree. */ return write_dentry_tree_recursive(root, p); } + + +static int +init_wimlib_dentry(struct wimlib_dir_entry *wdentry, + struct wim_dentry *dentry, + const WIMStruct *wim, + int flags) +{ + int ret; + size_t dummy; + const struct wim_inode *inode = dentry->d_inode; + struct wim_lookup_table_entry *lte; + const u8 *hash; + +#if TCHAR_IS_UTF16LE + wdentry->filename = dentry->file_name; + wdentry->dos_name = dentry->short_name; +#else + if (dentry_has_long_name(dentry)) { + ret = utf16le_to_tstr(dentry->file_name, + dentry->file_name_nbytes, + (tchar**)&wdentry->filename, + &dummy); + if (ret) + return ret; + } + if (dentry_has_short_name(dentry)) { + ret = utf16le_to_tstr(dentry->short_name, + dentry->short_name_nbytes, + (tchar**)&wdentry->dos_name, + &dummy); + if (ret) + return ret; + } +#endif + ret = calculate_dentry_full_path(dentry); + if (ret) + return ret; + wdentry->full_path = dentry->_full_path; + + for (struct wim_dentry *d = dentry; !dentry_is_root(d); d = d->parent) + wdentry->depth++; + + if (inode->i_security_id >= 0) { + const struct wim_security_data *sd = wim_const_security_data(wim); + wdentry->security_descriptor = sd->descriptors[inode->i_security_id]; + wdentry->security_descriptor_size = sd->sizes[inode->i_security_id]; + } + wdentry->reparse_tag = inode->i_reparse_tag; + wdentry->num_links = inode->i_nlink; + wdentry->attributes = inode->i_attributes; + wdentry->hard_link_group_id = inode->i_ino; + wdentry->creation_time = wim_timestamp_to_timespec(inode->i_creation_time); + wdentry->last_write_time = wim_timestamp_to_timespec(inode->i_last_write_time); + wdentry->last_access_time = wim_timestamp_to_timespec(inode->i_last_access_time); + + lte = inode_unnamed_lte(inode, wim->lookup_table); + if (lte) { + lte_to_wimlib_resource_entry(lte, &wdentry->streams[0].resource); + } else if (!is_zero_hash(hash = inode_unnamed_stream_hash(inode))) { + if (flags & WIMLIB_ITERATE_DIR_TREE_FLAG_RESOURCES_NEEDED) + return resource_not_found_error(inode, hash); + copy_hash(wdentry->streams[0].resource.sha1_hash, hash); + wdentry->streams[0].resource.is_missing = 1; + } + + for (unsigned i = 0; i < inode->i_num_ads; i++) { + if (!ads_entry_is_named_stream(&inode->i_ads_entries[i])) + continue; + lte = inode_stream_lte(inode, i + 1, wim->lookup_table); + wdentry->num_named_streams++; + if (lte) { + lte_to_wimlib_resource_entry(lte, &wdentry->streams[ + wdentry->num_named_streams].resource); + } else if (!is_zero_hash(hash = inode_stream_hash(inode, i + 1))) { + if (flags & WIMLIB_ITERATE_DIR_TREE_FLAG_RESOURCES_NEEDED) + return resource_not_found_error(inode, hash); + copy_hash(wdentry->streams[ + wdentry->num_named_streams].resource.sha1_hash, hash); + wdentry->streams[ + wdentry->num_named_streams].resource.is_missing = 1; + } + #if TCHAR_IS_UTF16LE + wdentry->streams[wdentry->num_named_streams].stream_name = + inode->i_ads_entries[i].stream_name; + #else + size_t dummy; + + ret = utf16le_to_tstr(inode->i_ads_entries[i].stream_name, + inode->i_ads_entries[i].stream_name_nbytes, + (tchar**)&wdentry->streams[ + wdentry->num_named_streams].stream_name, + &dummy); + if (ret) + return ret; + #endif + } + return 0; +} + +static void +free_wimlib_dentry(struct wimlib_dir_entry *wdentry) +{ +#if !TCHAR_IS_UTF16LE + FREE((tchar*)wdentry->filename); + FREE((tchar*)wdentry->dos_name); + for (unsigned i = 1; i <= wdentry->num_named_streams; i++) + FREE((tchar*)wdentry->streams[i].stream_name); +#endif + FREE(wdentry); +} + +struct iterate_dir_tree_ctx { + WIMStruct *wim; + int flags; + wimlib_iterate_dir_tree_callback_t cb; + void *user_ctx; +}; + +static int +do_iterate_dir_tree(WIMStruct *wim, + struct wim_dentry *dentry, int flags, + wimlib_iterate_dir_tree_callback_t cb, + void *user_ctx); + +static int +call_do_iterate_dir_tree(struct wim_dentry *dentry, void *_ctx) +{ + struct iterate_dir_tree_ctx *ctx = _ctx; + return do_iterate_dir_tree(ctx->wim, dentry, ctx->flags, + ctx->cb, ctx->user_ctx); +} + +static int +do_iterate_dir_tree(WIMStruct *wim, + struct wim_dentry *dentry, int flags, + wimlib_iterate_dir_tree_callback_t cb, + void *user_ctx) +{ + struct wimlib_dir_entry *wdentry; + int ret = WIMLIB_ERR_NOMEM; + + + wdentry = CALLOC(1, sizeof(struct wimlib_dir_entry) + + (1 + dentry->d_inode->i_num_ads) * + sizeof(struct wimlib_stream_entry)); + if (wdentry == NULL) + goto out; + + ret = init_wimlib_dentry(wdentry, dentry, wim, flags); + if (ret) + goto out_free_wimlib_dentry; + + if (!(flags & WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN)) { + ret = (*cb)(wdentry, user_ctx); + if (ret) + goto out_free_wimlib_dentry; + } + + if (flags & (WIMLIB_ITERATE_DIR_TREE_FLAG_RECURSIVE | + WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN)) + { + struct iterate_dir_tree_ctx ctx = { + .wim = wim, + .flags = flags &= ~WIMLIB_ITERATE_DIR_TREE_FLAG_CHILDREN, + .cb = cb, + .user_ctx = user_ctx, + }; + ret = for_dentry_child(dentry, call_do_iterate_dir_tree, &ctx); + } +out_free_wimlib_dentry: + free_wimlib_dentry(wdentry); +out: + return ret; +} + +struct image_iterate_dir_tree_ctx { + const tchar *path; + int flags; + wimlib_iterate_dir_tree_callback_t cb; + void *user_ctx; +}; + + +static int +image_do_iterate_dir_tree(WIMStruct *wim) +{ + struct image_iterate_dir_tree_ctx *ctx = wim->private; + struct wim_dentry *dentry; + + dentry = get_dentry(wim, ctx->path, WIMLIB_CASE_PLATFORM_DEFAULT); + if (dentry == NULL) + return WIMLIB_ERR_PATH_DOES_NOT_EXIST; + return do_iterate_dir_tree(wim, dentry, ctx->flags, ctx->cb, ctx->user_ctx); +} + +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_iterate_dir_tree(WIMStruct *wim, int image, const tchar *_path, + int flags, + wimlib_iterate_dir_tree_callback_t cb, void *user_ctx) +{ + tchar *path; + int ret; + + path = canonicalize_wim_path(_path); + if (path == NULL) + return WIMLIB_ERR_NOMEM; + struct image_iterate_dir_tree_ctx ctx = { + .path = path, + .flags = flags, + .cb = cb, + .user_ctx = user_ctx, + }; + wim->private = &ctx; + ret = for_image(wim, image, image_do_iterate_dir_tree); + FREE(path); + return ret; +} + +/* Returns %true iff the metadata of @inode and @template_inode are reasonably + * consistent with them being the same, unmodified file. */ +static bool +inode_metadata_consistent(const struct wim_inode *inode, + const struct wim_inode *template_inode, + const struct wim_lookup_table *template_lookup_table) +{ + /* Must have exact same creation time and last write time. */ + if (inode->i_creation_time != template_inode->i_creation_time || + inode->i_last_write_time != template_inode->i_last_write_time) + return false; + + /* Last access time may have stayed the same or increased, but certainly + * shouldn't have decreased. */ + if (inode->i_last_access_time < template_inode->i_last_access_time) + return false; + + /* Must have same number of alternate data stream entries. */ + if (inode->i_num_ads != template_inode->i_num_ads) + return false; + + /* If the stream entries for the inode are for some reason not resolved, + * then the hashes are already available and the point of this function + * is defeated. */ + if (!inode->i_resolved) + return false; + + /* Iterate through each stream and do some more checks. */ + for (unsigned i = 0; i <= inode->i_num_ads; i++) { + const struct wim_lookup_table_entry *lte, *template_lte; + + lte = inode_stream_lte_resolved(inode, i); + template_lte = inode_stream_lte(template_inode, i, + template_lookup_table); + + /* Compare stream sizes. */ + if (lte && template_lte) { + if (lte->size != template_lte->size) + return false; + + /* If hash happens to be available, compare with template. */ + if (!lte->unhashed && !template_lte->unhashed && + !hashes_equal(lte->hash, template_lte->hash)) + return false; + + } else if (lte && lte->size) { + return false; + } else if (template_lte && template_lte->size) { + return false; + } + } + + /* All right, barring a full checksum and given that the inodes share a + * path and the user isn't trying to trick us, these inodes most likely + * refer to the same file. */ + return true; +} + +/** + * Given an inode @inode that has been determined to be "the same" as another + * inode @template_inode in either the same WIM or another WIM, retrieve some + * useful stream information (e.g. checksums) from @template_inode. + * + * This assumes that the streams for @inode have been resolved (to point + * directly to the appropriate `struct wim_lookup_table_entry's) but do not + * necessarily have checksum information filled in. + */ +static int +inode_copy_checksums(struct wim_inode *inode, + struct wim_inode *template_inode, + WIMStruct *wim, + WIMStruct *template_wim) +{ + for (unsigned i = 0; i <= inode->i_num_ads; i++) { + struct wim_lookup_table_entry *lte, *template_lte; + struct wim_lookup_table_entry *replace_lte; + + lte = inode_stream_lte_resolved(inode, i); + template_lte = inode_stream_lte(template_inode, i, + template_wim->lookup_table); + + /* Only take action if both entries exist, the entry for @inode + * has no checksum calculated, but the entry for @template_inode + * does. */ + if (lte == NULL || template_lte == NULL || + !lte->unhashed || template_lte->unhashed) + continue; + + wimlib_assert(lte->refcnt == inode->i_nlink); + + /* If the WIM of the template image is the same as the WIM of + * the new image, then @template_lte can be used directly. + * + * Otherwise, look for a stream with the same hash in the WIM of + * the new image. If found, use it; otherwise re-use the entry + * being discarded, filling in the hash. */ + + if (wim == template_wim) + replace_lte = template_lte; + else + replace_lte = lookup_resource(wim->lookup_table, + template_lte->hash); + + list_del(<e->unhashed_list); + if (replace_lte) { + free_lookup_table_entry(lte); + } else { + copy_hash(lte->hash, template_lte->hash); + lte->unhashed = 0; + lookup_table_insert(wim->lookup_table, lte); + lte->refcnt = 0; + replace_lte = lte; + } + + if (i == 0) + inode->i_lte = replace_lte; + else + inode->i_ads_entries[i - 1].lte = replace_lte; + + replace_lte->refcnt += inode->i_nlink; + } + return 0; +} + +struct reference_template_args { + WIMStruct *wim; + WIMStruct *template_wim; +}; + +static int +dentry_reference_template(struct wim_dentry *dentry, void *_args) +{ + int ret; + struct wim_dentry *template_dentry; + struct wim_inode *inode, *template_inode; + struct reference_template_args *args = _args; + WIMStruct *wim = args->wim; + WIMStruct *template_wim = args->template_wim; + + if (dentry->d_inode->i_visited) + return 0; + + ret = calculate_dentry_full_path(dentry); + if (ret) + return ret; + + template_dentry = get_dentry(template_wim, dentry->_full_path, + WIMLIB_CASE_SENSITIVE); + if (template_dentry == NULL) { + DEBUG("\"%"TS"\": newly added file", dentry->_full_path); + return 0; + } + + inode = dentry->d_inode; + template_inode = template_dentry->d_inode; + + if (inode_metadata_consistent(inode, template_inode, + template_wim->lookup_table)) { + /*DEBUG("\"%"TS"\": No change detected", dentry->_full_path);*/ + ret = inode_copy_checksums(inode, template_inode, + wim, template_wim); + inode->i_visited = 1; + } else { + DEBUG("\"%"TS"\": change detected!", dentry->_full_path); + ret = 0; + } + return ret; +} + +/* API function documented in wimlib.h */ +WIMLIBAPI int +wimlib_reference_template_image(WIMStruct *wim, int new_image, + WIMStruct *template_wim, int template_image, + int flags, wimlib_progress_func_t progress_func) +{ + int ret; + struct wim_image_metadata *new_imd; + + if (wim == NULL || template_wim == NULL) + return WIMLIB_ERR_INVALID_PARAM; + + if (wim == template_wim && new_image == template_image) + return WIMLIB_ERR_INVALID_PARAM; + + if (new_image < 1 || new_image > wim->hdr.image_count) + return WIMLIB_ERR_INVALID_IMAGE; + + if (!wim_has_metadata(wim)) + return WIMLIB_ERR_METADATA_NOT_FOUND; + + new_imd = wim->image_metadata[new_image - 1]; + if (!new_imd->modified) + return WIMLIB_ERR_INVALID_PARAM; + + ret = select_wim_image(template_wim, template_image); + if (ret) + return ret; + + struct reference_template_args args = { + .wim = wim, + .template_wim = template_wim, + }; + + ret = for_dentry_in_tree(new_imd->root_dentry, + dentry_reference_template, &args); + dentry_tree_clear_inode_visited(new_imd->root_dentry); + return ret; +}