]> wimlib.net Git - wimlib/blobdiff - src/ntfs-capture.c
capture_ntfs_streams(): Set 'ret' properly
[wimlib] / src / ntfs-capture.c
index b3f79eb4f4ce43139f234ad49582b52161c0240b..f447468eb6b044a91cb3ab72c7232ec0bcac86ad 100644 (file)
@@ -6,7 +6,7 @@
  */
 
 /*
- * Copyright (C) 2012 Eric Biggers
+ * Copyright (C) 2012, 2013 Eric Biggers
  *
  * This file is part of wimlib, a library for working with WIM files.
  *
 #include <ntfs-3g/endians.h>
 #include <ntfs-3g/types.h>
 
-#include "wimlib_internal.h"
-
-
+#include "buffer_io.h"
 #include "dentry.h"
 #include "lookup_table.h"
-#include "buffer_io.h"
+#include "security.h"
+#include "wimlib_internal.h"
+
 #include <ntfs-3g/layout.h>
 #include <ntfs-3g/acls.h>
 #include <ntfs-3g/attrib.h>
 #include <ntfs-3g/misc.h>
 #include <ntfs-3g/reparse.h>
-#include <ntfs-3g/security.h> /* security.h before xattrs.h */
+#include <ntfs-3g/security.h> /* ntfs-3g/security.h before ntfs-3g/xattrs.h */
 #include <ntfs-3g/xattrs.h>
 #include <ntfs-3g/volume.h>
 #include <stdlib.h>
 #include <unistd.h>
 #include <errno.h>
-#include "rbtree.h"
-
-/* Red-black tree that maps SHA1 message digests of security descriptors to
- * security IDs, which are themselves indices into the table of security
- * descriptors in the 'struct wim_security_data'. */
-struct sd_set {
-       struct wim_security_data *sd;
-       struct rb_root rb_root;
-};
 
-struct sd_node {
-       int security_id;
-       u8 hash[SHA1_HASH_SIZE];
-       struct rb_node rb_node;
-};
+#ifdef HAVE_ALLOCA_H
+#include <alloca.h>
+#endif
 
-static void free_sd_tree(struct rb_node *node)
-{
-       if (node) {
-               free_sd_tree(node->rb_left);
-               free_sd_tree(node->rb_right);
-               FREE(container_of(node, struct sd_node, rb_node));
-       }
-}
-/* Frees a security descriptor index set. */
-static void destroy_sd_set(struct sd_set *sd_set)
+static inline ntfschar *
+attr_record_name(ATTR_RECORD *ar)
 {
-       free_sd_tree(sd_set->rb_root.rb_node);
+       return (ntfschar*)((u8*)ar + le16_to_cpu(ar->name_offset));
 }
 
-/* Inserts a a new node into the security descriptor index tree. */
-static void insert_sd_node(struct sd_set *set, struct sd_node *new)
+static ntfs_attr *
+open_ntfs_attr(ntfs_inode *ni, struct ntfs_location *loc)
 {
-       struct rb_root *root = &set->rb_root;
-       struct rb_node **p = &(root->rb_node);
-       struct rb_node *rb_parent = NULL;
-
-       while (*p) {
-               struct sd_node *this = container_of(*p, struct sd_node, rb_node);
-               int cmp = hashes_cmp(new->hash, this->hash);
+       ntfs_attr *na;
 
-               rb_parent = *p;
-               if (cmp < 0)
-                       p = &((*p)->rb_left);
-               else if (cmp > 0)
-                       p = &((*p)->rb_right);
-               else
-                       wimlib_assert(0); /* Duplicate SHA1 message digest */
+       na = ntfs_attr_open(ni,
+                           loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
+                           loc->stream_name,
+                           loc->stream_name_nchars);
+       if (!na) {
+               ERROR_WITH_ERRNO("Failed to open attribute of \"%"TS"\" in "
+                                "NTFS volume", loc->path);
        }
-       rb_link_node(&new->rb_node, rb_parent, p);
-       rb_insert_color(&new->rb_node, root);
+       return na;
 }
 
-/* Returns the index of the security descriptor having a SHA1 message digest of
- * @hash.  If not found, return -1. */
-static int lookup_sd(struct sd_set *set, const u8 hash[SHA1_HASH_SIZE])
+int
+read_ntfs_file_prefix(const struct wim_lookup_table_entry *lte,
+                     u64 size,
+                     consume_data_callback_t cb,
+                     void *ctx_or_buf,
+                     int _ignored_flags)
 {
-       struct rb_node *node = set->rb_root.rb_node;
+       struct ntfs_location *loc = lte->ntfs_loc;
+       ntfs_volume *vol = loc->ntfs_vol;
+       ntfs_inode *ni;
+       ntfs_attr *na;
+       s64 pos;
+       s64 bytes_remaining;
+       void *out_buf;
+       int ret;
 
-       while (node) {
-               struct sd_node *sd_node = container_of(node, struct sd_node, rb_node);
-               int cmp = hashes_cmp(hash, sd_node->hash);
-               if (cmp < 0)
-                       node = node->rb_left;
-               else if (cmp > 0)
-                       node = node->rb_right;
-               else
-                       return sd_node->security_id;
+       ni = ntfs_pathname_to_inode(vol, NULL, loc->path);
+       if (!ni) {
+               ERROR_WITH_ERRNO("Can't find NTFS inode for \"%"TS"\"", loc->path);
+               ret = WIMLIB_ERR_NTFS_3G;
+               goto out;
        }
-       return -1;
-}
 
-/*
- * Adds a security descriptor to the indexed security descriptor set as well as
- * the corresponding `struct wim_security_data', and returns the new security
- * ID; or, if there is an existing security descriptor that is the same, return
- * the security ID for it.  If a new security descriptor cannot be allocated,
- * return -1.
- */
-static int sd_set_add_sd(struct sd_set *sd_set, const char descriptor[],
-                        size_t size)
-{
-       u8 hash[SHA1_HASH_SIZE];
-       int security_id;
-       struct sd_node *new;
-       u8 **descriptors;
-       u64 *sizes;
-       u8 *descr_copy;
-       struct wim_security_data *sd;
-
-       sha1_buffer((const u8*)descriptor, size, hash);
-
-       security_id = lookup_sd(sd_set, hash);
-       if (security_id >= 0) /* Identical descriptor already exists */
-               return security_id;
-
-       /* Need to add a new security descriptor */
-       new = MALLOC(sizeof(*new));
-       if (!new)
-               goto out;
-       descr_copy = MALLOC(size);
-       if (!descr_copy)
-               goto out_free_node;
-
-       sd = sd_set->sd;
-
-       memcpy(descr_copy, descriptor, size);
-       new->security_id = sd->num_entries;
-       copy_hash(new->hash, hash);
-
-       descriptors = REALLOC(sd->descriptors,
-                             (sd->num_entries + 1) * sizeof(sd->descriptors[0]));
-       if (!descriptors)
-               goto out_free_descr;
-       sd->descriptors = descriptors;
-       sizes = REALLOC(sd->sizes,
-                       (sd->num_entries + 1) * sizeof(sd->sizes[0]));
-       if (!sizes)
-               goto out_free_descr;
-       sd->sizes = sizes;
-       sd->descriptors[sd->num_entries] = descr_copy;
-       sd->sizes[sd->num_entries] = size;
-       sd->num_entries++;
-       DEBUG("There are now %d security descriptors", sd->num_entries);
-       sd->total_length += size + sizeof(sd->sizes[0]);
-       insert_sd_node(sd_set, new);
-       return new->security_id;
-out_free_descr:
-       FREE(descr_copy);
-out_free_node:
-       FREE(new);
-out:
-       return -1;
-}
+       na = open_ntfs_attr(ni, loc);
+       if (!na) {
+               ret = WIMLIB_ERR_NTFS_3G;
+               goto out_close_ntfs_inode;
+       }
 
-static inline ntfschar *attr_record_name(ATTR_RECORD *ar)
-{
-       return (ntfschar*)((u8*)ar + le16_to_cpu(ar->name_offset));
+       if (cb)
+               out_buf = alloca(WIM_CHUNK_SIZE);
+       else
+               out_buf = ctx_or_buf;
+       pos = (loc->is_reparse_point) ? 8 : 0;
+       bytes_remaining = size;
+       while (bytes_remaining) {
+               s64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
+               if (ntfs_attr_pread(na, pos, to_read, out_buf) != to_read) {
+                       ERROR_WITH_ERRNO("Error reading \"%"TS"\"", loc->path);
+                       ret = WIMLIB_ERR_NTFS_3G;
+                       goto out_close_ntfs_attr;
+               }
+               pos += to_read;
+               bytes_remaining -= to_read;
+               if (cb) {
+                       ret = cb(out_buf, to_read, ctx_or_buf);
+                       if (ret)
+                               goto out_close_ntfs_attr;
+               } else {
+                       out_buf += to_read;
+               }
+       }
+       ret = 0;
+out_close_ntfs_attr:
+       ntfs_attr_close(na);
+out_close_ntfs_inode:
+       ntfs_inode_close(ni);
+out:
+       return ret;
 }
 
-/* Calculates the SHA1 message digest of a NTFS attribute.
- *
- * @ni:  The NTFS inode containing the attribute.
- * @ar:         The ATTR_RECORD describing the attribute.
- * @md:  If successful, the returned SHA1 message digest.
- * @reparse_tag_ret:   Optional pointer into which the first 4 bytes of the
- *                             attribute will be written (to get the reparse
- *                             point ID)
- *
- * Return 0 on success or nonzero on error.
- */
-static int ntfs_attr_sha1sum(ntfs_inode *ni, ATTR_RECORD *ar,
-                            u8 md[SHA1_HASH_SIZE],
-                            bool is_reparse_point,
-                            u32 *reparse_tag_ret)
+static int
+read_reparse_tag(ntfs_inode *ni, struct ntfs_location *loc,
+                u32 *reparse_tag_ret)
 {
-       s64 pos = 0;
-       s64 bytes_remaining;
-       char buf[BUFFER_SIZE];
+       int ret;
+       u8 buf[8];
        ntfs_attr *na;
-       SHA_CTX ctx;
 
-       na = ntfs_attr_open(ni, ar->type, attr_record_name(ar),
-                           ar->name_length);
+       na = open_ntfs_attr(ni, loc);
        if (!na) {
-               ERROR_WITH_ERRNO("Failed to open NTFS attribute");
-               return WIMLIB_ERR_NTFS_3G;
-       }
-
-       bytes_remaining = na->data_size;
-
-       if (is_reparse_point) {
-               if (ntfs_attr_pread(na, 0, 8, buf) != 8)
-                       goto out_error;
-               *reparse_tag_ret = le32_to_cpu(*(u32*)buf);
-               pos = 8;
-               bytes_remaining -= 8;
+               ret = WIMLIB_ERR_NTFS_3G;
+               goto out;
        }
 
-       sha1_init(&ctx);
-       while (bytes_remaining) {
-               s64 to_read = min(bytes_remaining, sizeof(buf));
-               if (ntfs_attr_pread(na, pos, to_read, buf) != to_read)
-                       goto out_error;
-               sha1_update(&ctx, buf, to_read);
-               pos += to_read;
-               bytes_remaining -= to_read;
+       if (ntfs_attr_pread(na, 0, 8, buf) != 8) {
+               ERROR_WITH_ERRNO("Error reading reparse data");
+               ret = WIMLIB_ERR_NTFS_3G;
+               goto out_close_ntfs_attr;
        }
-       sha1_final(md, &ctx);
+       *reparse_tag_ret = le32_to_cpu(*(u32*)buf);
+       DEBUG("ReparseTag = %#x", *reparse_tag_ret);
+       ret = 0;
+out_close_ntfs_attr:
        ntfs_attr_close(na);
-       return 0;
-out_error:
-       ERROR_WITH_ERRNO("Error reading NTFS attribute");
-       return WIMLIB_ERR_NTFS_3G;
+out:
+       return ret;
+
 }
 
 /* Load the streams from a file or reparse point in the NTFS volume into the WIM
  * lookup table */
-static int capture_ntfs_streams(struct wim_dentry *dentry, ntfs_inode *ni,
-                               char path[], size_t path_len,
-                               struct wim_lookup_table *lookup_table,
-                               ntfs_volume **ntfs_vol_p,
-                               ATTR_TYPES type)
+static int
+capture_ntfs_streams(struct wim_inode *inode,
+                    ntfs_inode *ni,
+                    char *path,
+                    size_t path_len,
+                    struct wim_lookup_table *lookup_table,
+                    ntfs_volume *vol,
+                    ATTR_TYPES type)
 {
        ntfs_attr_search_ctx *actx;
-       u8 attr_hash[SHA1_HASH_SIZE];
-       struct ntfs_location *ntfs_loc = NULL;
-       int ret = 0;
+       struct ntfs_location *ntfs_loc;
+       int ret;
        struct wim_lookup_table_entry *lte;
 
        DEBUG2("Capturing NTFS data streams from `%s'", path);
@@ -271,131 +194,111 @@ static int capture_ntfs_streams(struct wim_dentry *dentry, ntfs_inode *ni,
        while (!ntfs_attr_lookup(type, NULL, 0,
                                 CASE_SENSITIVE, 0, NULL, 0, actx))
        {
-               char *stream_name_utf8;
-               u32 reparse_tag;
                u64 data_size = ntfs_get_attribute_value_length(actx->attr);
                u64 name_length = actx->attr->name_length;
+               u32 stream_id;
 
                if (data_size == 0) {
-                       if (errno != 0) {
-                               ERROR_WITH_ERRNO("Failed to get size of attribute of "
-                                                "`%s'", path);
-                               ret = WIMLIB_ERR_NTFS_3G;
-                               goto out_put_actx;
-                       }
                        /* Empty stream.  No lookup table entry is needed. */
                        lte = NULL;
+                       ntfs_loc = NULL;
                } else {
-                       if (type == AT_REPARSE_POINT && data_size < 8) {
-                               ERROR("`%s': reparse point buffer too small",
-                                     path);
-                               ret = WIMLIB_ERR_NTFS_3G;
+                       ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
+                       if (!ntfs_loc) {
+                               ret = WIMLIB_ERR_NOMEM;
                                goto out_put_actx;
                        }
-                       /* Checksum the stream. */
-                       ret = ntfs_attr_sha1sum(ni, actx->attr, attr_hash,
-                                               type == AT_REPARSE_POINT, &reparse_tag);
-                       if (ret != 0)
-                               goto out_put_actx;
-
-                       /* Make a lookup table entry for the stream, or use an existing
-                        * one if there's already an identical stream. */
-                       lte = __lookup_resource(lookup_table, attr_hash);
-                       ret = WIMLIB_ERR_NOMEM;
-                       if (lte) {
-                               lte->refcnt++;
-                       } else {
-                               ntfs_loc = CALLOC(1, sizeof(*ntfs_loc));
-                               if (!ntfs_loc)
-                                       goto out_put_actx;
-                               ntfs_loc->ntfs_vol_p = ntfs_vol_p;
-                               ntfs_loc->path_utf8 = MALLOC(path_len + 1);
-                               if (!ntfs_loc->path_utf8)
+                       ntfs_loc->ntfs_vol = vol;
+                       ntfs_loc->path = MALLOC(path_len + 1);
+                       if (!ntfs_loc->path) {
+                               ret = WIMLIB_ERR_NOMEM;
+                               goto out_free_ntfs_loc;
+                       }
+                       memcpy(ntfs_loc->path, path, path_len + 1);
+                       if (name_length) {
+                               ntfs_loc->stream_name = MALLOC(name_length * 2);
+                               if (!ntfs_loc->stream_name) {
+                                       ret = WIMLIB_ERR_NOMEM;
                                        goto out_free_ntfs_loc;
-                               memcpy(ntfs_loc->path_utf8, path, path_len + 1);
-                               if (name_length) {
-                                       ntfs_loc->stream_name_utf16 = MALLOC(name_length * 2);
-                                       if (!ntfs_loc->stream_name_utf16)
-                                               goto out_free_ntfs_loc;
-                                       memcpy(ntfs_loc->stream_name_utf16,
-                                              attr_record_name(actx->attr),
-                                              actx->attr->name_length * 2);
-                                       ntfs_loc->stream_name_utf16_num_chars = name_length;
                                }
+                               memcpy(ntfs_loc->stream_name,
+                                      attr_record_name(actx->attr),
+                                      actx->attr->name_length * 2);
+                               ntfs_loc->stream_name_nchars = name_length;
+                       }
 
-                               lte = new_lookup_table_entry();
-                               if (!lte)
-                                       goto out_free_ntfs_loc;
-                               lte->ntfs_loc = ntfs_loc;
-                               lte->resource_location = RESOURCE_IN_NTFS_VOLUME;
-                               if (type == AT_REPARSE_POINT) {
-                                       dentry->d_inode->i_reparse_tag = reparse_tag;
-                                       ntfs_loc->is_reparse_point = true;
-                                       lte->resource_entry.original_size = data_size - 8;
-                                       lte->resource_entry.size = data_size - 8;
-                               } else {
-                                       ntfs_loc->is_reparse_point = false;
-                                       lte->resource_entry.original_size = data_size;
-                                       lte->resource_entry.size = data_size;
+                       lte = new_lookup_table_entry();
+                       if (!lte) {
+                               ret = WIMLIB_ERR_NOMEM;
+                               goto out_free_ntfs_loc;
+                       }
+                       lte->resource_location = RESOURCE_IN_NTFS_VOLUME;
+                       lte->ntfs_loc = ntfs_loc;
+                       ntfs_loc = NULL;
+                       if (type == AT_REPARSE_POINT) {
+                               if (data_size < 8) {
+                                       ERROR("Invalid reparse data (only %u bytes)!",
+                                             (unsigned)data_size);
+                                       ret = WIMLIB_ERR_NTFS_3G;
+                                       goto out_free_lte;
                                }
-                               ntfs_loc = NULL;
-                               DEBUG("Add resource for `%s' (size = %"PRIu64")",
-                                     dentry->file_name_utf8,
-                                     lte->resource_entry.original_size);
-                               copy_hash(lte->hash, attr_hash);
-                               lookup_table_insert(lookup_table, lte);
+                               lte->ntfs_loc->is_reparse_point = true;
+                               lte->resource_entry.original_size = data_size - 8;
+                               ret = read_reparse_tag(ni, lte->ntfs_loc,
+                                                      &inode->i_reparse_tag);
+                               if (ret)
+                                       goto out_free_lte;
+                       } else {
+                               lte->ntfs_loc->is_reparse_point = false;
+                               lte->resource_entry.original_size = data_size;
                        }
                }
                if (name_length == 0) {
                        /* Unnamed data stream.  Put the reference to it in the
                         * dentry's inode. */
-               #if 0
-                       if (dentry->d_inode->i_lte) {
-                               ERROR("Found two un-named data streams for "
-                                     "`%s'", path);
+                       if (inode->i_lte) {
+                               ERROR("Found two un-named data streams for `%s'",
+                                     path);
                                ret = WIMLIB_ERR_NTFS_3G;
                                goto out_free_lte;
                        }
-                       dentry->d_inode->i_lte = lte;
-               #else
-                       if (dentry->d_inode->i_lte) {
-                               WARNING("Found two un-named data streams for "
-                                       "`%s'", path);
-                               free_lookup_table_entry(lte);
-                       } else {
-                               dentry->d_inode->i_lte = lte;
-                       }
-               #endif
+                       stream_id = 0;
+                       inode->i_lte = lte;
                } else {
                        /* Named data stream.  Put the reference to it in the
                         * alternate data stream entries */
                        struct wim_ads_entry *new_ads_entry;
-                       size_t stream_name_utf8_len;
 
-                       ret = utf16_to_utf8((const char*)attr_record_name(actx->attr),
-                                           name_length * 2,
-                                           &stream_name_utf8,
-                                           &stream_name_utf8_len);
-                       if (ret != 0)
-                               goto out_free_lte;
-                       new_ads_entry = inode_add_ads(dentry->d_inode, stream_name_utf8);
-                       FREE(stream_name_utf8);
-                       if (!new_ads_entry)
+                       new_ads_entry = inode_add_ads_utf16le(inode,
+                                                             attr_record_name(actx->attr),
+                                                             name_length * 2);
+                       if (!new_ads_entry) {
+                               ret = WIMLIB_ERR_NOMEM;
                                goto out_free_lte;
-
-                       wimlib_assert(new_ads_entry->stream_name_len == name_length * 2);
-
+                       }
+                       wimlib_assert(new_ads_entry->stream_name_nbytes == name_length * 2);
+                       stream_id = new_ads_entry->stream_id;
                        new_ads_entry->lte = lte;
                }
+               if (lte) {
+                       lookup_table_insert_unhashed(lookup_table, lte,
+                                                    inode, stream_id);
+               }
+       }
+       if (errno == ENOENT) {
+               ret = 0;
+       } else {
+               ERROR_WITH_ERRNO("Error listing NTFS attributes from `%s'",
+                                path);
+               ret = WIMLIB_ERR_NTFS_3G;
        }
-       ret = 0;
        goto out_put_actx;
 out_free_lte:
        free_lookup_table_entry(lte);
 out_free_ntfs_loc:
        if (ntfs_loc) {
-               FREE(ntfs_loc->path_utf8);
-               FREE(ntfs_loc->stream_name_utf16);
+               FREE(ntfs_loc->path);
+               FREE(ntfs_loc->stream_name);
                FREE(ntfs_loc);
        }
 out_put_actx:
@@ -403,7 +306,7 @@ out_put_actx:
        if (ret == 0)
                DEBUG2("Successfully captured NTFS streams from `%s'", path);
        else
-               ERROR("Failed to capture NTFS streams from `%s", path);
+               ERROR("Failed to capture NTFS streams from `%s'", path);
        return ret;
 }
 
@@ -415,33 +318,33 @@ struct dos_name_map {
 struct dos_name_node {
        struct rb_node rb_node;
        char dos_name[24];
-       int name_len_bytes;
+       int name_nbytes;
        u64 ntfs_ino;
 };
 
 /* Inserts a new DOS name into the map */
-static int insert_dos_name(struct dos_name_map *map,
-                          const ntfschar *dos_name, int name_len,
-                          u64 ntfs_ino)
+static int
+insert_dos_name(struct dos_name_map *map, const ntfschar *dos_name,
+               size_t name_nbytes, u64 ntfs_ino)
 {
        struct dos_name_node *new_node;
        struct rb_node **p;
        struct rb_root *root;
        struct rb_node *rb_parent;
 
-       DEBUG("DOS name_len = %d", name_len);
+       DEBUG("DOS name_len = %zu", name_nbytes);
        new_node = MALLOC(sizeof(struct dos_name_node));
        if (!new_node)
                return -1;
 
        /* DOS names are supposed to be 12 characters max (that's 24 bytes,
         * assuming 2-byte ntfs characters) */
-       wimlib_assert(name_len * sizeof(ntfschar) <= sizeof(new_node->dos_name));
+       wimlib_assert(name_nbytes <= sizeof(new_node->dos_name));
 
        /* Initialize the DOS name, DOS name length, and NTFS inode number of
         * the red-black tree node */
-       memcpy(new_node->dos_name, dos_name, name_len * sizeof(ntfschar));
-       new_node->name_len_bytes = name_len * sizeof(ntfschar);
+       memcpy(new_node->dos_name, dos_name, name_nbytes);
+       new_node->name_nbytes = name_nbytes;
        new_node->ntfs_ino = ntfs_ino;
 
        /* Insert the red-black tree node */
@@ -491,7 +394,8 @@ lookup_dos_name(const struct dos_name_map *map, u64 ntfs_ino)
        return NULL;
 }
 
-static int set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
+static int
+set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
 {
        const struct dos_name_map *map = arg;
        const struct dos_name_node *node;
@@ -499,12 +403,13 @@ static int set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
        if (dentry->is_win32_name) {
                node = lookup_dos_name(map, dentry->d_inode->i_ino);
                if (node) {
-                       dentry->short_name = MALLOC(node->name_len_bytes);
+                       dentry->short_name = MALLOC(node->name_nbytes + 2);
                        if (!dentry->short_name)
                                return WIMLIB_ERR_NOMEM;
                        memcpy(dentry->short_name, node->dos_name,
-                              node->name_len_bytes);
-                       dentry->short_name_len = node->name_len_bytes;
+                              node->name_nbytes);
+                       dentry->short_name[node->name_nbytes / 2] = 0;
+                       dentry->short_name_nbytes = node->name_nbytes;
                        DEBUG("Assigned DOS name to ino %"PRIu64,
                              dentry->d_inode->i_ino);
                } else {
@@ -516,7 +421,8 @@ static int set_dentry_dos_name(struct wim_dentry *dentry, void *arg)
        return 0;
 }
 
-static void free_dos_name_tree(struct rb_node *node) {
+static void
+free_dos_name_tree(struct rb_node *node) {
        if (node) {
                free_dos_name_tree(node->rb_left);
                free_dos_name_tree(node->rb_right);
@@ -524,7 +430,8 @@ static void free_dos_name_tree(struct rb_node *node) {
        }
 }
 
-static void destroy_dos_name_map(struct dos_name_map *map)
+static void
+destroy_dos_name_map(struct dos_name_map *map)
 {
        free_dos_name_tree(map->rb_root.rb_node);
 }
@@ -535,62 +442,69 @@ struct readdir_ctx {
        char *path;
        size_t path_len;
        struct wim_lookup_table *lookup_table;
+       struct wim_inode_table *inode_table;
        struct sd_set *sd_set;
        struct dos_name_map *dos_name_map;
-       const struct capture_config *config;
-       ntfs_volume **ntfs_vol_p;
+       const struct wimlib_capture_config *config;
+       ntfs_volume *vol;
        int add_image_flags;
        wimlib_progress_func_t progress_func;
 };
 
 static int
-build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p, ntfs_inode *dir_ni,
-                                ntfs_inode *ni, char path[], size_t path_len,
+build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
+                                ntfs_inode *dir_ni,
+                                ntfs_inode *ni,
+                                char *path,
+                                size_t path_len,
                                 int name_type,
                                 struct wim_lookup_table *lookup_table,
+                                struct wim_inode_table *inode_table,
                                 struct sd_set *sd_set,
-                                const struct capture_config *config,
-                                ntfs_volume **ntfs_vol_p,
+                                const struct wimlib_capture_config *config,
+                                ntfs_volume *ntfs_vol,
                                 int add_image_flags,
                                 wimlib_progress_func_t progress_func);
 
-static int wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
-                                   const int name_len, const int name_type,
-                                   const s64 pos, const MFT_REF mref,
-                                   const unsigned dt_type)
+static int
+wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
+                        const int name_nchars, const int name_type,
+                        const s64 pos, const MFT_REF mref,
+                        const unsigned dt_type)
 {
        struct readdir_ctx *ctx;
-       size_t utf8_name_len;
-       char *utf8_name;
+       size_t mbs_name_nbytes;
+       char *mbs_name;
        struct wim_dentry *child;
        int ret;
        size_t path_len;
+       size_t name_nbytes = name_nchars * sizeof(ntfschar);
 
        ctx = dirent;
        if (name_type & FILE_NAME_DOS) {
                /* If this is the entry for a DOS name, store it for later. */
                ret = insert_dos_name(ctx->dos_name_map, name,
-                                     name_len, mref & MFT_REF_MASK_CPU);
+                                     name_nbytes, mref & MFT_REF_MASK_CPU);
 
                /* Return now if an error occurred or if this is just a DOS name
                 * and not a Win32+DOS name. */
                if (ret != 0 || name_type == FILE_NAME_DOS)
-                       return ret;
+                       goto out;
        }
-       ret = utf16_to_utf8((const char*)name, name_len * 2,
-                           &utf8_name, &utf8_name_len);
-       if (ret != 0)
-               return -1;
+       ret = utf16le_to_tstr(name, name_nbytes,
+                             &mbs_name, &mbs_name_nbytes);
+       if (ret)
+               goto out;
 
-       if (utf8_name[0] == '.' &&
-            (utf8_name[1] == '\0' ||
-             (utf8_name[1] == '.' && utf8_name[2] == '\0'))) {
+       if (mbs_name[0] == '.' &&
+            (mbs_name[1] == '\0' ||
+             (mbs_name[1] == '.' && mbs_name[2] == '\0'))) {
                /* . or .. entries
                 *
                 * note: name_type is POSIX for these, so DOS names will not
                 * have been inserted for them.  */
                ret = 0;
-               goto out_free_utf8_name;
+               goto out_free_mbs_name;
        }
 
        /* Open the inode for this directory entry and recursively capture the
@@ -598,25 +512,29 @@ static int wim_ntfs_capture_filldir(void *dirent, const ntfschar *name,
        ntfs_inode *ni = ntfs_inode_open(ctx->dir_ni->vol, mref);
        if (!ni) {
                ERROR_WITH_ERRNO("Failed to open NTFS inode");
-               goto out_free_utf8_name;
+               ret = -1;
+               goto out_free_mbs_name;
        }
        path_len = ctx->path_len;
        if (path_len != 1)
                ctx->path[path_len++] = '/';
-       memcpy(ctx->path + path_len, utf8_name, utf8_name_len + 1);
-       path_len += utf8_name_len;
+       memcpy(ctx->path + path_len, mbs_name, mbs_name_nbytes + 1);
+       path_len += mbs_name_nbytes;
        child = NULL;
        ret = build_dentry_tree_ntfs_recursive(&child, ctx->dir_ni,
                                               ni, ctx->path, path_len, name_type,
-                                              ctx->lookup_table, ctx->sd_set,
-                                              ctx->config, ctx->ntfs_vol_p,
+                                              ctx->lookup_table,
+                                              ctx->inode_table,
+                                              ctx->sd_set,
+                                              ctx->config, ctx->vol,
                                               ctx->add_image_flags,
                                               ctx->progress_func);
        if (child)
                dentry_add_child(ctx->parent, child);
        ntfs_inode_close(ni);
-out_free_utf8_name:
-       FREE(utf8_name);
+out_free_mbs_name:
+       FREE(mbs_name);
+out:
        return ret;
 }
 
@@ -624,27 +542,31 @@ out_free_utf8_name:
  * At the same time, update the WIM lookup table with lookup table entries for
  * the NTFS streams, and build an array of security descriptors.
  */
-static int build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
-                                           ntfs_inode *dir_ni,
-                                           ntfs_inode *ni,
-                                           char path[],
-                                           size_t path_len,
-                                           int name_type,
-                                           struct wim_lookup_table *lookup_table,
-                                           struct sd_set *sd_set,
-                                           const struct capture_config *config,
-                                           ntfs_volume **ntfs_vol_p,
-                                           int add_image_flags,
-                                           wimlib_progress_func_t progress_func)
+static int
+build_dentry_tree_ntfs_recursive(struct wim_dentry **root_ret,
+                                ntfs_inode *dir_ni,
+                                ntfs_inode *ni,
+                                char *path,
+                                size_t path_len,
+                                int name_type,
+                                struct wim_lookup_table *lookup_table,
+                                struct wim_inode_table *inode_table,
+                                struct sd_set *sd_set,
+                                const struct wimlib_capture_config *config,
+                                ntfs_volume *vol,
+                                int add_image_flags,
+                                wimlib_progress_func_t progress_func)
 {
        u32 attributes;
        int ret;
        struct wim_dentry *root;
+       struct wim_inode *inode;
+       ATTR_TYPES stream_type;
 
-       if (exclude_path(path, config, false)) {
+       if (exclude_path(path, path_len, config, false)) {
                /* Exclude a file or directory tree based on the capture
                 * configuration file */
-               if ((add_image_flags & WIMLIB_ADD_IMAGE_FLAG_VERBOSE)
+               if ((add_image_flags & WIMLIB_ADD_IMAGE_FLAG_EXCLUDE_VERBOSE)
                    && progress_func)
                {
                        union wimlib_progress_info info;
@@ -652,14 +574,15 @@ static int build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
                        info.scan.excluded = true;
                        progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
                }
-               *root_p = NULL;
-               return 0;
+               root = NULL;
+               ret = 0;
+               goto out;
        }
 
        /* Get file attributes */
        struct SECURITY_CONTEXT ctx;
        memset(&ctx, 0, sizeof(ctx));
-       ctx.vol = ni->vol;
+       ctx.vol = vol;
        ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ATTRIB,
                                         ni, dir_ni, (char *)&attributes,
                                         sizeof(u32));
@@ -678,35 +601,48 @@ static int build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
                progress_func(WIMLIB_PROGRESS_MSG_SCAN_DENTRY, &info);
        }
 
-       /* Create the new WIM dentry */
-       root = new_dentry_with_timeless_inode(path_basename(path));
-       if (!root) {
-               if (errno == EILSEQ)
-                       return WIMLIB_ERR_INVALID_UTF8_STRING;
-               else if (errno == ENOMEM)
-                       return WIMLIB_ERR_NOMEM;
-               else
-                       return WIMLIB_ERR_ICONV_NOT_AVAILABLE;
-       }
-       *root_p = root;
+       /* Create a WIM dentry with an associated inode, which may be shared */
+       ret = inode_table_new_dentry(inode_table,
+                                    path_basename_with_len(path, path_len),
+                                    ni->mft_no,
+                                    0,
+                                    &root);
+       if (ret)
+               return ret;
+
+       inode = root->d_inode;
+
+       if (inode->i_nlink > 1) /* Shared inode; nothing more to do */
+               goto out;
 
-       if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name */
+       if (name_type & FILE_NAME_WIN32) /* Win32 or Win32+DOS name (rather than POSIX) */
                root->is_win32_name = 1;
-       root->d_inode->i_creation_time    = le64_to_cpu(ni->creation_time);
-       root->d_inode->i_last_write_time  = le64_to_cpu(ni->last_data_change_time);
-       root->d_inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
-       root->d_inode->i_attributes       = le32_to_cpu(attributes);
-       root->d_inode->i_ino              = ni->mft_no;
-       root->d_inode->i_resolved         = 1;
-
-       if (attributes & FILE_ATTR_REPARSE_POINT) {
-               /* Junction point, symbolic link, or other reparse point */
-               ret = capture_ntfs_streams(root, ni, path, path_len,
-                                          lookup_table, ntfs_vol_p,
-                                          AT_REPARSE_POINT);
-       } else if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
-
-               /* Normal directory */
+       inode->i_creation_time    = le64_to_cpu(ni->creation_time);
+       inode->i_last_write_time  = le64_to_cpu(ni->last_data_change_time);
+       inode->i_last_access_time = le64_to_cpu(ni->last_access_time);
+       inode->i_attributes       = le32_to_cpu(attributes);
+       inode->i_resolved         = 1;
+
+       if (attributes & FILE_ATTR_REPARSE_POINT)
+               stream_type = AT_REPARSE_POINT;
+       else
+               stream_type = AT_DATA;
+
+       /* Capture the file's streams; more specifically, this is supposed to:
+        *
+        * - Regular files: capture unnamed data stream and any named data
+        *   streams
+        * - Directories: capture any named data streams
+        * - Reparse points: capture reparse data only
+        */
+       ret = capture_ntfs_streams(inode, ni, path, path_len, lookup_table,
+                                  vol, stream_type);
+       if (ret)
+               goto out;
+
+       if (ni->mrec->flags & MFT_RECORD_IS_DIRECTORY) {
+
+               /* Recurse to directory children */
                s64 pos = 0;
                struct dos_name_map dos_name_map = { .rb_root = {.rb_node = NULL} };
                struct readdir_ctx ctx = {
@@ -715,10 +651,11 @@ static int build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
                        .path            = path,
                        .path_len        = path_len,
                        .lookup_table    = lookup_table,
+                       .inode_table     = inode_table,
                        .sd_set          = sd_set,
                        .dos_name_map    = &dos_name_map,
                        .config          = config,
-                       .ntfs_vol_p      = ntfs_vol_p,
+                       .vol             = vol,
                        .add_image_flags = add_image_flags,
                        .progress_func   = progress_func,
                };
@@ -731,64 +668,75 @@ static int build_dentry_tree_ntfs_recursive(struct wim_dentry **root_p,
                                               &dos_name_map);
                }
                destroy_dos_name_map(&dos_name_map);
-       } else {
-               /* Normal file */
-               ret = capture_ntfs_streams(root, ni, path, path_len,
-                                          lookup_table, ntfs_vol_p,
-                                          AT_DATA);
+               if (ret)
+                       goto out;
        }
-       if (ret != 0)
-               return ret;
 
-       /* Get security descriptor */
-       char _sd[1];
-       char *sd = _sd;
-       errno = 0;
-       ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ACL,
-                                        ni, dir_ni, sd,
-                                        sizeof(sd));
-       if (ret > sizeof(sd)) {
-               sd = alloca(ret);
+       if (!(add_image_flags & WIMLIB_ADD_IMAGE_FLAG_NO_ACLS)) {
+               /* Get security descriptor */
+               char _sd[1];
+               char *sd = _sd;
+               errno = 0;
                ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ACL,
-                                                ni, dir_ni, sd, ret);
-       }
-       if (ret > 0) {
-               root->d_inode->i_security_id = sd_set_add_sd(sd_set, sd, ret);
-               if (root->d_inode->i_security_id == -1) {
-                       ERROR("Out of memory");
-                       return WIMLIB_ERR_NOMEM;
+                                                ni, dir_ni, sd,
+                                                sizeof(sd));
+               if (ret > sizeof(sd)) {
+                       sd = alloca(ret);
+                       ret = ntfs_xattr_system_getxattr(&ctx, XATTR_NTFS_ACL,
+                                                        ni, dir_ni, sd, ret);
+               }
+               if (ret > 0) {
+                       inode->i_security_id = sd_set_add_sd(sd_set, sd, ret);
+                       if (inode->i_security_id == -1) {
+                               ERROR("Out of memory");
+                               ret = WIMLIB_ERR_NOMEM;
+                               goto out;
+                       }
+                       DEBUG("Added security ID = %u for `%s'",
+                             inode->i_security_id, path);
+                       ret = 0;
+               } else if (ret < 0) {
+                       ERROR_WITH_ERRNO("Failed to get security information from "
+                                        "`%s'", path);
+                       ret = WIMLIB_ERR_NTFS_3G;
+               } else {
+                       inode->i_security_id = -1;
+                       DEBUG("No security ID for `%s'", path);
                }
-               DEBUG("Added security ID = %u for `%s'",
-                     root->d_inode->i_security_id, path);
-               ret = 0;
-       } else if (ret < 0) {
-               ERROR_WITH_ERRNO("Failed to get security information from "
-                                "`%s'", path);
-               ret = WIMLIB_ERR_NTFS_3G;
-       } else {
-               root->d_inode->i_security_id = -1;
-               DEBUG("No security ID for `%s'", path);
        }
+out:
+       if (ret == 0)
+               *root_ret = root;
+       else
+               free_dentry_tree(root, lookup_table);
        return ret;
 }
 
-int build_dentry_tree_ntfs(struct wim_dentry **root_p,
-                          const char *device,
-                          struct wim_lookup_table *lookup_table,
-                          struct wim_security_data *sd,
-                          const struct capture_config *config,
-                          int add_image_flags,
-                          wimlib_progress_func_t progress_func,
-                          void *extra_arg)
+
+int
+do_ntfs_umount(struct _ntfs_volume *vol)
+{
+       DEBUG("Unmounting NTFS volume");
+       if (ntfs_umount(vol, FALSE))
+               return WIMLIB_ERR_NTFS_3G;
+       else
+               return 0;
+}
+
+int
+build_dentry_tree_ntfs(struct wim_dentry **root_p,
+                      const char *device,
+                      struct wim_lookup_table *lookup_table,
+                      struct wim_inode_table *inode_table,
+                      struct sd_set *sd_set,
+                      const struct wimlib_capture_config *config,
+                      int add_image_flags,
+                      wimlib_progress_func_t progress_func,
+                      void *extra_arg)
 {
        ntfs_volume *vol;
        ntfs_inode *root_ni;
        int ret;
-       struct sd_set sd_set = {
-               .sd = sd,
-               .rb_root = {NULL},
-       };
-       ntfs_volume **ntfs_vol_p = extra_arg;
 
        DEBUG("Mounting NTFS volume `%s' read-only", device);
 
@@ -820,8 +768,8 @@ int build_dentry_tree_ntfs(struct wim_dentry **root_p,
                goto out;
        }
 
-       /* Currently we assume that all the UTF-8 paths fit into this length and
-        * there is no check for overflow. */
+       /* Currently we assume that all the paths fit into this length and there
+        * is no check for overflow. */
        char *path = MALLOC(32768);
        if (!path) {
                ERROR("Could not allocate memory for NTFS pathname");
@@ -833,21 +781,21 @@ int build_dentry_tree_ntfs(struct wim_dentry **root_p,
        path[1] = '\0';
        ret = build_dentry_tree_ntfs_recursive(root_p, NULL, root_ni, path, 1,
                                               FILE_NAME_POSIX, lookup_table,
-                                              &sd_set,
-                                              config, ntfs_vol_p,
+                                              inode_table,
+                                              sd_set,
+                                              config, vol,
                                               add_image_flags,
                                               progress_func);
 out_cleanup:
        FREE(path);
        ntfs_inode_close(root_ni);
-       destroy_sd_set(&sd_set);
 out:
        ntfs_index_ctx_put(vol->secure_xsii);
        ntfs_index_ctx_put(vol->secure_xsdh);
        ntfs_inode_close(vol->secure_ni);
 
        if (ret) {
-               if (ntfs_umount(vol, FALSE) != 0) {
+               if (do_ntfs_umount(vol)) {
                        ERROR_WITH_ERRNO("Failed to unmount NTFS volume `%s'",
                                         device);
                        if (ret == 0)
@@ -856,7 +804,7 @@ out:
        } else {
                /* We need to leave the NTFS volume mounted so that we can read
                 * the NTFS files again when we are actually writing the WIM */
-               *ntfs_vol_p = vol;
+               *(ntfs_volume**)extra_arg = vol;
        }
        return ret;
 }