#include "util.h"
#include "config.h"
#include "list.h"
+#include "sha1.h"
+#include "rbtree.h"
#include <string.h>
+#ifdef WITH_FUSE
+#include <pthread.h>
+#endif
struct stat;
-struct lookup_table;
-typedef struct WIMStruct WIMStruct;
-
-/* Size of the struct dentry up to and including the file_name_len. */
+struct wim_lookup_table;
+struct WIMStruct;
+struct wim_lookup_table_entry;
+struct wimfs_fd;
+struct wim_inode;
+struct wim_dentry;
+
+/* Size of the struct wim_dentry up to and including the file_name_len. */
#define WIM_DENTRY_DISK_SIZE 102
+/* Size of on-disk WIM alternate data stream entry, in bytes, up to and
+ * including the stream length field (see below). */
#define WIM_ADS_ENTRY_DISK_SIZE 38
-#ifndef WIM_HASH_SIZE
-#define WIM_HASH_SIZE 20
-#endif
-
-/*
- * Reparse tags documented at
+/*
+ * Reparse tags documented at
* http://msdn.microsoft.com/en-us/library/dd541667(v=prot.10).aspx
*/
#define WIM_IO_REPARSE_TAG_RESERVED_ZERO 0x00000000
#define FILE_ATTRIBUTE_ENCRYPTED 0x00004000
#define FILE_ATTRIBUTE_VIRTUAL 0x00010000
-/* Alternate data stream entry */
-struct ads_entry {
- /* SHA-1 message digest of stream contents */
- u8 hash[WIM_HASH_SIZE];
- /* Length of stream name (UTF-16) */
+/* Alternate data stream entry.
+ *
+ * We read this from disk in the read_ads_entries() function; see that function
+ * for more explanation. */
+struct wim_ads_entry {
+ union {
+ /* SHA-1 message digest of stream contents */
+ u8 hash[SHA1_HASH_SIZE];
+
+ /* The corresponding lookup table entry (only for resolved
+ * streams) */
+ struct wim_lookup_table_entry *lte;
+ };
+
+ /* Length of stream name (UTF-16). This is in bytes, not characters,
+ * and does not include the terminating null character */
u16 stream_name_len;
/* Length of stream name (UTF-8) */
/* Stream name (UTF-8) */
char *stream_name_utf8;
-};
-static inline u64 ads_entry_length(const struct ads_entry *entry)
-{
- u64 len = WIM_ADS_ENTRY_DISK_SIZE + entry->stream_name_len + 2;
- return (len + 7) & ~7;
-}
+#ifdef WITH_FUSE
+ /* Number to identify an alternate data stream even after it's possibly
+ * been moved or renamed. */
+ u32 stream_id;
+#endif
+};
-static inline void destroy_ads_entry(struct ads_entry *entry)
-{
- FREE(entry->stream_name);
- FREE(entry->stream_name_utf8);
- memset(entry, 0, sizeof(entry));
-}
-static inline bool ads_entry_has_name(const struct ads_entry *entry,
- const char *name, size_t name_len)
+static inline bool ads_entries_have_same_name(const struct wim_ads_entry *entry_1,
+ const struct wim_ads_entry *entry_2)
{
- if (entry->stream_name_utf8_len != name_len)
+ if (entry_1->stream_name_len != entry_2->stream_name_len)
return false;
- return memcmp(entry->stream_name_utf8, name, name_len) == 0;
+ return memcmp(entry_1->stream_name, entry_2->stream_name,
+ entry_1->stream_name_len) == 0;
}
+/*
+ * In-memory structure for a WIM directory entry (dentry). There is a directory
+ * tree for each image in the WIM.
+ *
+ * Note that this is a directory entry and not an inode. Since NTFS allows hard
+ * links, it's possible for a NTFS inode to correspond to multiple WIM dentries.
+ * The hard link group ID field of the on-disk WIM dentry tells us the number of
+ * the NTFS inode that the dentry corresponds to (and this gets placed in
+ * d_inode->i_ino).
+ *
+ * Unfortunately, WIM files do not have an analogue to an inode; instead certain
+ * information, such as file attributes, the security descriptor, and file
+ * streams is replicated in each hard-linked dentry, even though this
+ * information really is associated with an inode. In-memory, we fix up this
+ * flaw by allocating a `struct wim_inode' for each dentry that contains some of
+ * this duplicated information, then combining the inodes for each hard link
+ * group together.
+ *
+ * Confusingly, it's possible for stream information to be missing from a dentry
+ * in a hard link set, in which case the stream information needs to be gotten
+ * from one of the other dentries in the hard link set. In addition, it is
+ * possible for dentries to have inconsistent security IDs, file attributes, or
+ * file streams when they share the same hard link ID (don't even ask. I hope
+ * that Microsoft may have fixed this problem, since I've only noticed it in the
+ * 'install.wim' for Windows 7). For those dentries, we have to use the
+ * conflicting fields to split up the hard link groups. (See
+ * dentry_tree_fix_inodes() in hardlink.c).
+ */
+struct wim_dentry {
+ /* Byte 0 */
-/* In-memory structure for a directory entry. There is a directory tree for
- * each image in the WIM. */
-struct dentry {
- /* The parent of this directory entry. */
- struct dentry *parent;
+ /* The inode for this dentry */
+ struct wim_inode *d_inode;
+
+ /* Byte 8 */
+
+ /* Red-black tree of sibling dentries */
+ struct rb_node rb_node;
+
+ /* Byte 32 */
+
+ /* Length of short filename, in bytes, not including the terminating
+ * zero wide-character. */
+ u16 short_name_len;
+
+ /* Length of file name, in bytes, not including the terminating zero
+ * wide-character. */
+ u16 file_name_len;
+
+ /* Length of the filename converted into UTF-8, in bytes, not including
+ * the terminating zero byte. */
+ u16 file_name_utf8_len;
- /* Linked list of sibling directory entries. */
- struct dentry *next;
+ u8 is_extracted : 1;
- struct dentry *prev;
+ /* Byte 40 */
- /* Pointer to a child of this directory entry. */
- struct dentry *children;
+ /* Pointer to the filename converted to UTF-8 (malloc()ed buffer). */
+ char *file_name_utf8;
+
+ /* Byte 48 */
+
+ struct list_head tmp_list;
+
+ /* Byte 64 */
+
+ /* List of dentries in the inode (hard link set) */
+ struct list_head d_alias;
- /* Size of directory entry, in bytes. Typical size is around 104 to 120
- * bytes. */
- /* It is possible for the length field to be 0. This situation, which
+ /* The parent of this directory entry. */
+ struct wim_dentry *parent;
+
+ /*
+ * Size of directory entry on disk, in bytes. Typical size is around
+ * 104 to 120 bytes.
+ *
+ * It is possible for the length 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. */
+ * included only the length field, but that takes up 8 bytes.
+ *
+ * The length here includes the base directory entry on disk as well as
+ * the long and short filenames. It does NOT include any alternate
+ * stream entries that may follow the directory entry, even though the
+ * size of those needs to be considered. The length SHOULD be 8-byte
+ * aligned, although we don't require it to be. We do require the
+ * length to be large enough to hold the file name(s) of the dentry;
+ * additionally, a warning is issued if this field is larger than the
+ * aligned size.
+ */
u64 length;
+ /* The offset, from the start of the uncompressed WIM metadata resource
+ * for this image, of this dentry's child dentries. 0 if the directory
+ * entry has no children, which is the case for regular files or reparse
+ * points. */
+ u64 subdir_offset;
- /* The file attributes associated with this file. */
- u32 attributes;
+ /* Number of references to the dentry tree itself, as in multiple
+ * WIMStructs */
+ u32 refcnt;
- /* The index of the node in the security table that contains this file's
- * security information. If -1, no security information exists for this
- * file. */
- int32_t security_id;
+ u32 full_path_utf8_len;
- /* The offset, from the start of the metadata section, of this directory
- * entry's child files. 0 if the directory entry has no children. */
- u64 subdir_offset;
+ /* Pointer to the UTF-16 short filename (malloc()ed buffer) */
+ char *short_name;
- /* Reserved for future disuse. Currently ignoring these fields. */
- u64 unused1;
- u64 unused2;
+ /* Pointer to the UTF-16 filename (malloc()ed buffer). */
+ char *file_name;
- /* Timestamps for the entry. The timestamps are the number of
- * 100-nanosecond intervals that have elapsed since 12:00 A.M., January
- * 1st, 1601, UTC. */
- u64 creation_time;
- u64 last_access_time;
- u64 last_write_time;
+ /* Full path (UTF-8) to this dentry (malloc()ed buffer). */
+ char *full_path_utf8;
+};
- /* A hash of the file's contents. */
- u8 hash[WIM_HASH_SIZE];
+#define rbnode_dentry(node) container_of(node, struct wim_dentry, rb_node)
+
+/*
+ * WIM inode.
+ *
+ * As mentioned above, in the WIM file that is no on-disk analogue of a real
+ * inode, as most of these fields are duplicated in the dentries.
+ */
+struct wim_inode {
+ /* Timestamps for the inode. The timestamps are the number of
+ * 100-nanosecond intervals that have elapsed since 12:00 A.M., January
+ * 1st, 1601, UTC. This is the same format used in NTFS inodes. */
+ u64 i_creation_time;
+ u64 i_last_access_time;
+ u64 i_last_write_time;
+
+ /* The file attributes associated with this inode. This is a bitwise OR
+ * of the FILE_ATTRIBUTE_* flags. */
+ u32 i_attributes;
+
+ /* The index of the security descriptor in the WIM image's table of
+ * security descriptors that contains this file's security information.
+ * If -1, no security information exists for this file. */
+ int32_t i_security_id;
+
+ /* %true iff the inode's lookup table entries has been resolved (i.e.
+ * the @lte field is valid, but the @hash field is not valid)
+ *
+ * (This is not an on-disk field.) */
+ u8 i_resolved : 1;
+
+ /* %true iff verify_inode() has run on this inode. */
+ u8 i_verified : 1;
+
+ /* Number of alternate data streams associated with this inode */
+ u16 i_num_ads;
+
+ /* A hash of the file's contents, or a pointer to the lookup table entry
+ * for this dentry if the lookup table entries have been resolved.
+ *
+ * More specifically, this is for the un-named default file stream, as
+ * opposed to the alternate (named) file streams, which may have their
+ * own lookup table entries. */
+ union {
+ u8 i_hash[SHA1_HASH_SIZE];
+ struct wim_lookup_table_entry *i_lte;
+ };
/* Identity of a reparse point. See
* http://msdn.microsoft.com/en-us/library/windows/desktop/aa365503(v=vs.85).aspx
* for what a reparse point is. */
- u32 reparse_tag;
+ u32 i_reparse_tag;
- /* Although M$'s documentation does not tell you this, it seems that the
- * reparse_reserved field does not actually exist. So the hard_link
- * field directly follows the reparse_tag on disk. EXCEPT when the
- * dentry is actually a reparse point... well, just take a look at the
- * read_dentry() function. */
- //u32 reparse_reserved;
+ /* Number of dentries that reference this inode */
+ u32 i_nlink;
- /* If the reparse_reserved field existed, there would be a 4-byte gap
- * here to align hard_link on an 8-byte field. However,
- * reparse_reserved does not actually exist, so there is no gap here. */
+ /* Alternate data stream entries. */
+ struct wim_ads_entry *i_ads_entries;
- /* If the file is part of a hard link set, all the directory entries in
- * the set will share the same value for this field. */
- u64 hard_link;
+ /* Inode number */
+ u64 i_ino;
- /* Number of alternate data streams associated with this file. */
- u16 num_ads;
+ /* List of dentries that reference this inode (there should be
+ * link_count of them) */
+ struct list_head i_dentry;
- /* Length of short filename, in bytes, not including the terminating
- * zero wide-character. */
- u16 short_name_len;
+ struct hlist_node i_hlist;
- /* Length of file name, in bytes, not including the terminating zero
- * wide-character. */
- u16 file_name_len;
+ struct list_head i_lte_inode_list;
- /* Length of the filename converted into UTF-8, in bytes, not including
- * the terminating zero byte. */
- u16 file_name_utf8_len;
+ char *i_extracted_file;
- /* Pointer to the short filename */
- char *short_name;
+ /* Root of a red-black tree storing the children of this inode (if
+ * non-empty, implies the inode is a directory, although that is also
+ * noted in the @attributes field.) */
+ struct rb_root i_children;
- /* Pointer to the filename. */
- char *file_name;
+#ifdef WITH_FUSE
+ /* wimfs file descriptors table for the inode */
+ u16 i_num_opened_fds;
+ u16 i_num_allocated_fds;
+ struct wimfs_fd **i_fds;
- /* Pointer to the filename converted to UTF-8. */
- char *file_name_utf8;
-
- /* Full path to this dentry. */
- char *full_path_utf8;
- u32 full_path_utf8_len;
+ /* Next alternate data stream ID to be assigned */
+ u32 i_next_stream_id;
- /* Alternate stream entries for this dentry. */
- struct ads_entry *ads_entries;
+ /* This mutex protects the inode's file descriptors table during
+ * read-only mounts. Read-write mounts are still restricted to 1
+ * thread. */
+ pthread_mutex_t i_mutex;
+#endif
+};
- /* Number of references to the dentry tree itself, as in multiple
- * WIMStructs */
- int refcnt;
+#define inode_for_each_dentry(dentry, inode) \
+ list_for_each_entry((dentry), &(inode)->i_dentry, d_alias)
- /* List of dentries in the hard link set */
- struct list_head link_group_list;
-};
+#define inode_add_dentry(dentry, inode) \
+ list_add_tail(&(dentry)->d_alias, &(inode)->i_dentry)
-/* Return hash of the "unnamed" (default) data stream. */
-static inline const u8 *dentry_hash(const struct dentry *dentry)
+static inline struct wim_dentry *inode_first_dentry(struct wim_inode *inode)
{
- /* If there are alternate data streams, the dentry hash field is zeroed
- * out, and we need to find the hash in the un-named data stream (should
- * be the first one, but check them in order just in case, and fall back
- * to the dentry hash field if we can't find an unnamed data stream). */
- for (u16 i = 0; i < dentry->num_ads; i++)
- if (dentry->ads_entries[i].stream_name_len == 0)
- return dentry->ads_entries[i].hash;
- return dentry->hash;
+ return container_of(inode->i_dentry.next, struct wim_dentry, d_alias);
}
-static inline size_t dentry_link_group_size(const struct dentry *dentry)
+static inline bool dentry_is_first_in_inode(const struct wim_dentry *dentry)
{
- size_t size = 0;
- struct list_head *list;
- list_for_each(list, &dentry->link_group_list)
- size++;
- if (size == 0)
- size = 1;
- return size;
+ return inode_first_dentry(dentry->d_inode) == dentry;
}
-extern struct ads_entry *dentry_get_ads_entry(struct dentry *dentry,
- const char *stream_name);
+extern u64 dentry_correct_total_length(const struct wim_dentry *dentry);
-extern struct ads_entry *dentry_add_ads(struct dentry *dentry,
- const char *stream_name);
+extern int for_dentry_in_tree(struct wim_dentry *root,
+ int (*visitor)(struct wim_dentry*, void*),
+ void *args);
-extern void dentry_remove_ads(struct dentry *dentry, struct ads_entry *entry);
+extern int for_dentry_in_rbtree(struct rb_node *node,
+ int (*visitor)(struct wim_dentry *, void *),
+ void *arg);
-extern const char *path_stream_name(const char *path);
+extern int for_dentry_in_tree_depth(struct wim_dentry *root,
+ int (*visitor)(struct wim_dentry*, void*),
+ void *args);
-extern u64 dentry_total_length(const struct dentry *dentry);
+extern int calculate_dentry_full_path(struct wim_dentry *dentry, void *ignore);
+extern void calculate_subdir_offsets(struct wim_dentry *dentry, u64 *subdir_offset_p);
+extern int set_dentry_name(struct wim_dentry *dentry, const char *new_name);
-extern void stbuf_to_dentry(const struct stat *stbuf, struct dentry *dentry);
+extern struct wim_dentry *get_dentry(struct WIMStruct *w, const char *path);
-extern void dentry_to_stbuf(const struct dentry *dentry, struct stat *stbuf,
- const struct lookup_table *table);
+extern struct wim_inode *wim_pathname_to_inode(struct WIMStruct *w,
+ const char *path);
-extern int for_dentry_in_tree(struct dentry *root,
- int (*visitor)(struct dentry*, void*),
- void *args);
+extern struct wim_dentry *
+get_dentry_child_with_name(const struct wim_dentry *dentry, const char *name);
-extern int for_dentry_in_tree_depth(struct dentry *root,
- int (*visitor)(struct dentry*, void*),
- void *args);
+extern struct wim_dentry *get_parent_dentry(struct WIMStruct *w,
+ const char *path);
-extern int calculate_dentry_full_path(struct dentry *dentry, void *ignore);
-extern void calculate_subdir_offsets(struct dentry *dentry, u64 *subdir_offset_p);
-extern int change_dentry_name(struct dentry *dentry, const char *new_name);
-extern int change_ads_name(struct ads_entry *entry, const char *new_name);
+extern int print_dentry(struct wim_dentry *dentry, void *lookup_table);
+extern int print_dentry_full_path(struct wim_dentry *entry, void *ignore);
-extern void unlink_dentry(struct dentry *dentry);
-extern void link_dentry(struct dentry *dentry, struct dentry *parent);
+extern struct wim_dentry *new_dentry(const char *name);
+extern struct wim_dentry *new_dentry_with_inode(const char *name);
+extern struct wim_dentry *new_dentry_with_timeless_inode(const char *name);
-extern int print_dentry(struct dentry *dentry, void *lookup_table);
-extern int print_dentry_full_path(struct dentry *entry, void *ignore);
+extern void free_inode(struct wim_inode *inode);
+extern void free_dentry(struct wim_dentry *dentry);
+extern void put_dentry(struct wim_dentry *dentry);
-extern struct dentry *get_dentry(WIMStruct *w, const char *path);
-extern struct dentry *get_parent_dentry(WIMStruct *w, const char *path);
-extern struct dentry *get_dentry_child_with_name(const struct dentry *dentry,
- const char *name);
-extern void dentry_update_all_timestamps(struct dentry *dentry);
-extern void init_dentry(struct dentry *dentry, const char *name);
-extern struct dentry *new_dentry(const char *name);
+extern void free_dentry_tree(struct wim_dentry *root,
+ struct wim_lookup_table *lookup_table);
+extern int increment_dentry_refcnt(struct wim_dentry *dentry, void *ignore);
-extern void free_dentry(struct dentry *dentry);
-extern void free_dentry_tree(struct dentry *root,
- struct lookup_table *lookup_table,
- bool lt_decrement_refcnt);
-extern int increment_dentry_refcnt(struct dentry *dentry, void *ignore);
-extern int decrement_dentry_refcnt(struct dentry *dentry, void *ignore);
+extern void unlink_dentry(struct wim_dentry *dentry);
+extern bool dentry_add_child(struct wim_dentry * restrict parent,
+ struct wim_dentry * restrict child);
-extern void calculate_dir_tree_statistics(struct dentry *root,
- struct lookup_table *table,
- u64 *dir_count_ret,
- u64 *file_count_ret,
- u64 *total_bytes_ret,
- u64 *hard_link_bytes_ret);
+extern struct wim_ads_entry *inode_get_ads_entry(struct wim_inode *inode,
+ const char *stream_name,
+ u16 *idx_ret);
-extern int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
- u64 offset, struct dentry *dentry);
+extern struct wim_ads_entry *inode_add_ads(struct wim_inode *dentry,
+ const char *stream_name);
-extern int read_dentry_tree(const u8 metadata_resource[],
- u64 metadata_resource_len, struct dentry *dentry);
+extern void inode_remove_ads(struct wim_inode *inode, u16 idx,
+ struct wim_lookup_table *lookup_table);
-extern u8 *write_dentry_tree(const struct dentry *tree, u8 *p);
+extern int read_dentry(const u8 metadata_resource[], u64 metadata_resource_len,
+ u64 offset, struct wim_dentry *dentry);
-extern int dentry_set_symlink_buf(struct dentry *dentry,
- const u8 symlink_buf_hash[]);
-/* Inline utility functions for WIMDentries */
+extern int read_dentry_tree(const u8 metadata_resource[],
+ u64 metadata_resource_len,
+ struct wim_dentry *dentry);
+extern u8 *write_dentry_tree(const struct wim_dentry *tree, u8 *p);
-static inline bool dentry_is_root(const struct dentry *dentry)
+static inline bool dentry_is_root(const struct wim_dentry *dentry)
{
return dentry->parent == dentry;
}
-static inline bool dentry_is_first_sibling(const struct dentry *dentry)
+static inline bool inode_is_directory(const struct wim_inode *inode)
{
- return dentry_is_root(dentry) || dentry->parent->children == dentry;
+ return (inode->i_attributes & FILE_ATTRIBUTE_DIRECTORY)
+ && !(inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT);
}
-static inline bool dentry_is_only_child(const struct dentry *dentry)
+static inline bool dentry_is_directory(const struct wim_dentry *dentry)
{
- return dentry->next == dentry;
+ return inode_is_directory(dentry->d_inode);
}
-static inline bool dentry_is_directory(const struct dentry *dentry)
+/* For our purposes, we consider "real" symlinks and "junction points" to both
+ * be symlinks. */
+static inline bool inode_is_symlink(const struct wim_inode *inode)
{
- return (dentry->attributes & FILE_ATTRIBUTE_DIRECTORY)
- && !(dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT);
+ return (inode->i_attributes & FILE_ATTRIBUTE_REPARSE_POINT)
+ && ((inode->i_reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK) ||
+ inode->i_reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT);
}
-/* For our purposes, we consider "real" symlinks and "junction points" to both
- * be symlinks. */
-static inline bool dentry_is_symlink(const struct dentry *dentry)
+static inline bool inode_is_regular_file(const struct wim_inode *inode)
+{
+ return !inode_is_directory(inode) && !inode_is_symlink(inode);
+}
+
+static inline bool dentry_is_regular_file(const struct wim_dentry *dentry)
{
- return (dentry->attributes & FILE_ATTRIBUTE_REPARSE_POINT)
- && ((dentry->reparse_tag == WIM_IO_REPARSE_TAG_SYMLINK) ||
- dentry->reparse_tag == WIM_IO_REPARSE_TAG_MOUNT_POINT);
+ return inode_is_regular_file(dentry->d_inode);
}
-static inline bool dentry_is_regular_file(const struct dentry *dentry)
+static inline bool inode_has_children(const struct wim_inode *inode)
{
- return !dentry_is_directory(dentry) && !dentry_is_symlink(dentry);
+ return inode->i_children.rb_node != NULL;
}
-static inline bool dentry_is_empty_directory(const struct dentry *dentry)
+static inline bool dentry_has_children(const struct wim_dentry *dentry)
{
- return dentry_is_directory(dentry) && dentry->children == NULL;
+ return inode_has_children(dentry->d_inode);
}
#endif