]> wimlib.net Git - wimlib/blobdiff - src/write.c
Fixes, comments
[wimlib] / src / write.c
index d985aafee2494b43eca7791cd72e8dd8b6ce9188..ccad7f07caa5c3f4b7af309995ab5974d2742aa0 100644 (file)
@@ -42,9 +42,6 @@
 #include "dentry.h"
 #include "lookup_table.h"
 #include "xml.h"
-#include "lzx.h"
-#include "xpress.h"
-
 
 #ifdef ENABLE_MULTITHREADED_COMPRESSION
 #  include <pthread.h>
@@ -152,8 +149,9 @@ out:
 
 /*
  * compress_func_t- Pointer to a function to compresses a chunk
- *                  of a WIM resource.  This may be either xpress_compress()
- *                  (xpress-compress.c) or lzx_compress() (lzx-compress.c).
+ *                  of a WIM resource.  This may be either
+ *                  wimlib_xpress_compress() (xpress-compress.c) or
+ *                  wimlib_lzx_compress() (lzx-compress.c).
  *
  * @chunk:       Uncompressed data of the chunk.
  * @chunk_size:          Size of the uncompressed chunk, in bytes.
@@ -179,9 +177,9 @@ compress_func_t
 get_compress_func(int out_ctype)
 {
        if (out_ctype == WIMLIB_COMPRESSION_TYPE_LZX)
-               return lzx_compress;
+               return wimlib_lzx_compress;
        else
-               return xpress_compress;
+               return wimlib_xpress_compress;
 }
 
 /*
@@ -189,22 +187,22 @@ get_compress_func(int out_ctype)
  *
  * @chunk:       Uncompressed data of the chunk.
  * @chunk_size:          Size of the chunk (<= WIM_CHUNK_SIZE)
- * @out_fp:      FILE * to write tho chunk to.
- * @out_ctype:   Compression type to use when writing the chunk (ignored if no
- *                     chunk table provided)
+ * @out_fp:      FILE * to write the chunk to.
+ * @compress:     Compression function to use (NULL if writing uncompressed
+ *                     data).
  * @chunk_tab:   Pointer to chunk table being created.  It is updated with the
  *                     offset of the chunk we write.
  *
  * Returns 0 on success; nonzero on failure.
  */
 static int
-write_wim_resource_chunk(const u8 chunk[], unsigned chunk_size,
+write_wim_resource_chunk(const void *chunk, unsigned chunk_size,
                         FILE *out_fp, compress_func_t compress,
                         struct chunk_table *chunk_tab)
 {
        const u8 *out_chunk;
        unsigned out_chunk_size;
-       if (chunk_tab) {
+       if (compress) {
                u8 *compressed_chunk = alloca(chunk_size);
 
                out_chunk_size = compress(chunk, chunk_size, compressed_chunk);
@@ -271,107 +269,6 @@ finish_wim_resource_chunk_tab(struct chunk_table *chunk_tab,
        return 0;
 }
 
-/* Prepare for multiple reads to a resource by caching a FILE * or NTFS
- * attribute pointer in the lookup table entry. */
-static int
-prepare_resource_for_read(struct wim_lookup_table_entry *lte
-
-                               #ifdef WITH_NTFS_3G
-                               , ntfs_inode **ni_ret
-                               #endif
-                       )
-{
-       switch (lte->resource_location) {
-       case RESOURCE_IN_FILE_ON_DISK:
-               if (!lte->file_on_disk_fp) {
-                       lte->file_on_disk_fp = tfopen(lte->file_on_disk, T("rb"));
-                       if (!lte->file_on_disk_fp) {
-                               ERROR_WITH_ERRNO("Failed to open the file "
-                                                "`%"TS"'", lte->file_on_disk);
-                               return WIMLIB_ERR_OPEN;
-                       }
-               }
-               break;
-#ifdef WITH_NTFS_3G
-       case RESOURCE_IN_NTFS_VOLUME:
-               if (!lte->attr) {
-                       struct ntfs_location *loc = lte->ntfs_loc;
-                       ntfs_inode *ni;
-                       wimlib_assert(loc);
-                       ni = ntfs_pathname_to_inode(*loc->ntfs_vol_p, NULL, loc->path);
-                       if (!ni) {
-                               ERROR_WITH_ERRNO("Failed to open inode `%"TS"' in NTFS "
-                                                "volume", loc->path);
-                               return WIMLIB_ERR_NTFS_3G;
-                       }
-                       lte->attr = ntfs_attr_open(ni,
-                                                  loc->is_reparse_point ? AT_REPARSE_POINT : AT_DATA,
-                                                  loc->stream_name,
-                                                  loc->stream_name_nchars);
-                       if (!lte->attr) {
-                               ERROR_WITH_ERRNO("Failed to open attribute of `%"TS"' in "
-                                                "NTFS volume", loc->path);
-                               ntfs_inode_close(ni);
-                               return WIMLIB_ERR_NTFS_3G;
-                       }
-                       *ni_ret = ni;
-               }
-               break;
-#endif
-#ifdef __WIN32__
-       case RESOURCE_WIN32:
-               if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
-                       lte->win32_file_on_disk_fp =
-                               win32_open_file_data_only(lte->file_on_disk);
-                       if (lte->win32_file_on_disk_fp == INVALID_HANDLE_VALUE) {
-                               ERROR("Win32 API: Can't open %"TS, lte->file_on_disk);
-                               win32_error_last();
-                               return WIMLIB_ERR_OPEN;
-                       }
-               }
-               break;
-#endif
-       default:
-               break;
-       }
-       return 0;
-}
-
-/* Undo prepare_resource_for_read() by closing the cached FILE * or NTFS
- * attribute. */
-static void
-end_wim_resource_read(struct wim_lookup_table_entry *lte
-                       #ifdef WITH_NTFS_3G
-                               , ntfs_inode *ni
-                       #endif
-                       )
-{
-       if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK
-           && lte->file_on_disk_fp)
-       {
-               fclose(lte->file_on_disk_fp);
-               lte->file_on_disk_fp = NULL;
-       }
-#ifdef WITH_NTFS_3G
-       else if (lte->resource_location == RESOURCE_IN_NTFS_VOLUME) {
-               if (lte->attr) {
-                       ntfs_attr_close(lte->attr);
-                       lte->attr = NULL;
-               }
-               if (ni)
-                       ntfs_inode_close(ni);
-       }
-#endif
-#ifdef __WIN32__
-       else if (lte->resource_location == RESOURCE_WIN32
-                && lte->win32_file_on_disk_fp != INVALID_HANDLE_VALUE)
-       {
-               win32_close_file(lte->win32_file_on_disk_fp);
-               lte->win32_file_on_disk_fp = INVALID_HANDLE_VALUE;
-       }
-#endif
-}
-
 static int
 write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
                                         FILE *out_fp,
@@ -384,217 +281,157 @@ write_uncompressed_resource_and_truncate(struct wim_lookup_table_entry *lte,
                                 "output WIM file", file_offset);
                return WIMLIB_ERR_WRITE;
        }
-       ret = write_wim_resource(lte, out_fp, WIMLIB_COMPRESSION_TYPE_NONE,
-                                out_res_entry, 0);
-       if (ret != 0)
+       ret = write_wim_resource(lte, out_fp,
+                                WIMLIB_COMPRESSION_TYPE_NONE,
+                                out_res_entry,
+                                0);
+       if (ret)
                return ret;
 
        return fflush_and_ftruncate(out_fp,
                                    file_offset + wim_resource_size(lte));
 }
 
-/*
- * Writes a WIM resource to a FILE * opened for writing.  The resource may be
- * written uncompressed or compressed depending on the @out_ctype parameter.
- *
- * If by chance the resource compresses to more than the original size (this may
- * happen with random data or files than are pre-compressed), the resource is
- * instead written uncompressed (and this is reflected in the @out_res_entry by
- * removing the WIM_RESHDR_FLAG_COMPRESSED flag).
- *
- * @lte:       The lookup table entry for the WIM resource.
- * @out_fp:    The FILE * to write the resource to.
- * @out_ctype:  The compression type of the resource to write.  Note: if this is
- *                     the same as the compression type of the WIM resource we
- *                     need to read, we simply copy the data (i.e. we do not
- *                     uncompress it, then compress it again).
- * @out_res_entry:  If non-NULL, a resource entry that is filled in with the
- *                 offset, original size, compressed size, and compression flag
- *                 of the output resource.
- *
- * Returns 0 on success; nonzero on failure.
- */
+struct write_resource_ctx {
+       compress_func_t compress;
+       struct chunk_table *chunk_tab;
+       FILE *out_fp;
+       SHA_CTX sha_ctx;
+       bool doing_sha;
+};
+
+static int
+write_resource_cb(const void *chunk, size_t chunk_size, void *_ctx)
+{
+       struct write_resource_ctx *ctx = _ctx;
+
+       if (ctx->doing_sha)
+               sha1_update(&ctx->sha_ctx, chunk, chunk_size);
+
+       if (ctx->compress) {
+               return write_wim_resource_chunk(chunk, chunk_size,
+                                               ctx->out_fp, ctx->compress,
+                                               ctx->chunk_tab);
+       } else {
+               if (fwrite(chunk, 1, chunk_size, ctx->out_fp) != chunk_size) {
+                       ERROR_WITH_ERRNO("Error writing to output WIM");
+                       return WIMLIB_ERR_WRITE;
+               } else {
+                       return 0;
+               }
+       }
+}
+
 int
 write_wim_resource(struct wim_lookup_table_entry *lte,
                   FILE *out_fp, int out_ctype,
                   struct resource_entry *out_res_entry,
                   int flags)
 {
-       u64 bytes_remaining;
-       u64 original_size;
-       u64 old_compressed_size;
-       u64 new_compressed_size;
-       u64 offset;
+       struct write_resource_ctx write_ctx;
+       u64 new_size;
+       off_t offset;
        int ret;
-       struct chunk_table *chunk_tab = NULL;
-       bool raw;
-       off_t file_offset;
-       compress_func_t compress = NULL;
-#ifdef WITH_NTFS_3G
-       ntfs_inode *ni = NULL;
-#endif
-
-       wimlib_assert(lte);
 
-       /* Original size of the resource */
-       original_size = wim_resource_size(lte);
-
-       /* Compressed size of the resource (as it exists now) */
-       old_compressed_size = wim_resource_compressed_size(lte);
+       if (wim_resource_size(lte) == 0) {
+               /* Empty resource; nothing needs to be done, so just return
+                * success. */
+               return 0;
+       }
 
-       /* Current offset in output file */
-       file_offset = ftello(out_fp);
-       if (file_offset == -1) {
-               ERROR_WITH_ERRNO("Failed to get offset in output "
-                                "stream");
+       offset = ftello(out_fp);
+       if (offset == -1) {
+               ERROR_WITH_ERRNO("Can't get position in output WIM");
                return WIMLIB_ERR_WRITE;
        }
 
-       /* Are the compression types the same?  If so, do a raw copy (copy
-        * without decompressing and recompressing the data). */
-       raw = (wim_resource_compression_type(lte) == out_ctype
-              && out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
-              && !(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS));
+       /* Can we simply copy the compressed data without recompressing it? */
 
-       if (raw) {
+       if (!(flags & WIMLIB_RESOURCE_FLAG_RECOMPRESS) &&
+           lte->resource_location == RESOURCE_IN_WIM &&
+           wimlib_get_compression_type(lte->wim) == out_ctype)
+       {
                flags |= WIMLIB_RESOURCE_FLAG_RAW;
-               bytes_remaining = old_compressed_size;
+               write_ctx.doing_sha = false;
        } else {
-               flags &= ~WIMLIB_RESOURCE_FLAG_RAW;
-               bytes_remaining = original_size;
+               write_ctx.doing_sha = true;
+               sha1_init(&write_ctx.sha_ctx);
        }
 
-       /* Empty resource; nothing needs to be done, so just return success. */
-       if (bytes_remaining == 0)
-               return 0;
-
-       /* Buffer for reading chunks for the resource */
-       u8 buf[min(WIM_CHUNK_SIZE, bytes_remaining)];
-
-       /* If we are writing a compressed resource and not doing a raw copy, we
-        * need to initialize the chunk table */
-       if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE && !raw) {
-               ret = begin_wim_resource_chunk_tab(lte, out_fp, file_offset,
-                                                  &chunk_tab);
-               if (ret != 0)
-                       goto out;
-       }
-
-       /* If the WIM resource is in an external file, open a FILE * to it so we
-        * don't have to open a temporary one in read_wim_resource() for each
-        * chunk. */
-#ifdef WITH_NTFS_3G
-       ret = prepare_resource_for_read(lte, &ni);
-#else
-       ret = prepare_resource_for_read(lte);
-#endif
-       if (ret != 0)
-               goto out;
-
-       /* If we aren't doing a raw copy, we will compute the SHA1 message
-        * digest of the resource as we read it, and verify it's the same as the
-        * hash given in the lookup table entry once we've finished reading the
-        * resource. */
-       SHA_CTX ctx;
-       if (!raw) {
-               sha1_init(&ctx);
-               compress = get_compress_func(out_ctype);
-       }
-       offset = 0;
-
-       /* While there are still bytes remaining in the WIM resource, read a
-        * chunk of the resource, update SHA1, then write that chunk using the
-        * desired compression type. */
-       do {
-               u64 to_read = min(bytes_remaining, WIM_CHUNK_SIZE);
-               ret = read_wim_resource(lte, buf, to_read, offset, flags);
-               if (ret != 0)
-                       goto out_fclose;
-               if (!raw)
-                       sha1_update(&ctx, buf, to_read);
-               ret = write_wim_resource_chunk(buf, to_read, out_fp,
-                                              compress, chunk_tab);
-               if (ret != 0)
-                       goto out_fclose;
-               bytes_remaining -= to_read;
-               offset += to_read;
-       } while (bytes_remaining);
-
-       /* Raw copy:  The new compressed size is the same as the old compressed
-        * size
-        *
-        * Using WIMLIB_COMPRESSION_TYPE_NONE:  The new compressed size is the
-        * original size
-        *
-        * Using a different compression type:  Call
-        * finish_wim_resource_chunk_tab() and it will provide the new
-        * compressed size.
-        */
-       if (raw) {
-               new_compressed_size = old_compressed_size;
+       /* Initialize the chunk table and set the compression function if
+        * compressing the resource. */
+       if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE ||
+           (flags & WIMLIB_RESOURCE_FLAG_RAW)) {
+               write_ctx.compress = NULL;
+               write_ctx.chunk_tab = NULL;
        } else {
-               if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE)
-                       new_compressed_size = original_size;
-               else {
-                       ret = finish_wim_resource_chunk_tab(chunk_tab, out_fp,
-                                                           &new_compressed_size);
-                       if (ret != 0)
-                               goto out_fclose;
-               }
+               write_ctx.compress = get_compress_func(out_ctype);
+               ret = begin_wim_resource_chunk_tab(lte, out_fp,
+                                                  offset,
+                                                  &write_ctx.chunk_tab);
+               if (ret)
+                       return ret;
        }
 
-       /* Verify SHA1 message digest of the resource, unless we are doing a raw
-        * write (in which case we never even saw the uncompressed data).  Or,
-        * if the hash we had before is all 0's, just re-set it to be the new
-        * hash. */
-       if (!raw) {
+       /* Write the data */
+       write_ctx.out_fp = out_fp;
+       ret = read_resource_prefix(lte, wim_resource_size(lte),
+                                  write_resource_cb, &write_ctx, 0);
+
+       /* Verify SHA1 message digest of the resource, or set the hash for the
+        * first time. */
+       if (write_ctx.doing_sha) {
                u8 md[SHA1_HASH_SIZE];
-               sha1_final(md, &ctx);
-               if (is_zero_hash(lte->hash)) {
+               sha1_final(md, &write_ctx.sha_ctx);
+               if (lte->unhashed) {
                        copy_hash(lte->hash, md);
                } else if (!hashes_equal(md, lte->hash)) {
                        ERROR("WIM resource has incorrect hash!");
-                       if (lte->resource_location == RESOURCE_IN_FILE_ON_DISK) {
-                               ERROR("We were reading it from `%"TS"'; maybe "
+                       if (lte_filename_valid(lte)) {
+                               ERROR("We were reading it from \"%"TS"\"; maybe "
                                      "it changed while we were reading it.",
                                      lte->file_on_disk);
                        }
                        ret = WIMLIB_ERR_INVALID_RESOURCE_HASH;
-                       goto out_fclose;
+                       goto out_free_chunk_tab;
                }
        }
 
-       if (!raw && new_compressed_size >= original_size &&
-           out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
-       {
-               /* Oops!  We compressed the resource to larger than the original
-                * size.  Write the resource uncompressed instead. */
-               ret = write_uncompressed_resource_and_truncate(lte,
-                                                              out_fp,
-                                                              file_offset,
-                                                              out_res_entry);
-               if (ret != 0)
-                       goto out_fclose;
+       out_res_entry->flags = lte->resource_entry.flags;
+       out_res_entry->original_size = wim_resource_size(lte);
+       out_res_entry->offset = offset;
+       if (flags & WIMLIB_RESOURCE_FLAG_RAW) {
+               /* Doing a raw write:  The new compressed size is the same as
+                * the compressed size in the other WIM. */
+               new_size = lte->resource_entry.size;
+       } else if (out_ctype == WIMLIB_COMPRESSION_TYPE_NONE) {
+               /* Using WIMLIB_COMPRESSION_TYPE_NONE:  The new compressed size
+                * is the original size. */
+               new_size = lte->resource_entry.original_size;
+               out_res_entry->flags &= ~WIM_RESHDR_FLAG_COMPRESSED;
        } else {
-               if (out_res_entry) {
-                       out_res_entry->size          = new_compressed_size;
-                       out_res_entry->original_size = original_size;
-                       out_res_entry->offset        = file_offset;
-                       out_res_entry->flags         = lte->resource_entry.flags
-                                                       & ~WIM_RESHDR_FLAG_COMPRESSED;
-                       if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE)
-                               out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
+               /* Using a different compression type:  Call
+                * finish_wim_resource_chunk_tab() and it will provide the new
+                * compressed size. */
+               ret = finish_wim_resource_chunk_tab(write_ctx.chunk_tab, out_fp,
+                                                   &new_size);
+               if (ret)
+                       goto out_free_chunk_tab;
+               if (new_size >= wim_resource_size(lte)) {
+                       /* Oops!  We compressed the resource to larger than the original
+                        * size.  Write the resource uncompressed instead. */
+                       ret = write_uncompressed_resource_and_truncate(lte,
+                                                                      out_fp,
+                                                                      offset,
+                                                                      out_res_entry);
+                       goto out_free_chunk_tab;
                }
+               out_res_entry->flags |= WIM_RESHDR_FLAG_COMPRESSED;
        }
+       out_res_entry->size = new_size;
        ret = 0;
-out_fclose:
-#ifdef WITH_NTFS_3G
-       end_wim_resource_read(lte, ni);
-#else
-       end_wim_resource_read(lte);
-#endif
-out:
-       FREE(chunk_tab);
+out_free_chunk_tab:
+       FREE(write_ctx.chunk_tab);
        return ret;
 }
 
@@ -761,8 +598,35 @@ do_write_streams_progress(union wimlib_progress_info *progress,
        }
 }
 
+static int
+sha1_chunk(const void *buf, size_t len, void *ctx)
+{
+       sha1_update(ctx, buf, len);
+       return 0;
+}
+
+static int
+sha1_resource(struct wim_lookup_table_entry *lte)
+{
+       int ret;
+       SHA_CTX sha_ctx;
+
+       sha1_init(&sha_ctx);
+       ret = read_resource_prefix(lte, wim_resource_size(lte),
+                                  sha1_chunk, &sha_ctx, 0);
+       if (ret == 0)
+               sha1_final(lte->hash, &sha_ctx);
+       return ret;
+}
+
+enum {
+       STREAMS_MERGED = 0,
+       STREAMS_NOT_MERGED = 1,
+};
+
 static int
 do_write_stream_list(struct list_head *my_resources,
+                    struct wim_lookup_table *lookup_table,
                     FILE *out_fp,
                     int out_ctype,
                     wimlib_progress_func_t progress_func,
@@ -770,18 +634,61 @@ do_write_stream_list(struct list_head *my_resources,
                     int write_resource_flags)
 {
        int ret;
-       struct wim_lookup_table_entry *lte, *tmp;
+       struct wim_lookup_table_entry *lte;
+
+       while (!list_empty(my_resources)) {
+               lte = container_of(my_resources->next,
+                                  struct wim_lookup_table_entry,
+                                  write_streams_list);
+               list_del(&lte->write_streams_list);
+               if (lte->unhashed && !lte->unique_size) {
+                       struct wim_lookup_table_entry *duplicate_lte;
+                       struct wim_lookup_table_entry **my_ptr;
+
+                       my_ptr = lte->my_ptr;
+                       ret = sha1_resource(lte);
+                       if (ret)
+                               return ret;
+                       duplicate_lte = __lookup_resource(lookup_table, lte->hash);
+                       if (duplicate_lte) {
+                               bool new_stream = (duplicate_lte->out_refcnt == 0);
+                               duplicate_lte->refcnt += lte->refcnt;
+                               duplicate_lte->out_refcnt += lte->refcnt;
+                               *my_ptr = duplicate_lte;
+                               free_lookup_table_entry(lte);
+
+                               if (new_stream) {
+                                       lte = duplicate_lte;
+                                       DEBUG("Stream of length %"PRIu64" is duplicate "
+                                             "with one already in WIM",
+                                             wim_resource_size(duplicate_lte));
+                               } else {
+                                       DEBUG("Discarding duplicate stream of length %"PRIu64,
+                                             wim_resource_size(duplicate_lte));
+                                       goto skip_to_progress;
+                               }
+
+                       } else {
+                               lookup_table_insert(lookup_table, lte);
+                               lte->out_refcnt = lte->refcnt;
+                               lte->unhashed = 0;
+                       }
+               }
+
+               wimlib_assert(lte->out_refcnt != 0);
 
-       list_for_each_entry_safe(lte, tmp, my_resources, staging_list) {
                ret = write_wim_resource(lte,
                                         out_fp,
                                         out_ctype,
                                         &lte->output_resource_entry,
                                         write_resource_flags);
-               if (ret != 0)
+               if (ret)
                        return ret;
-               list_del(&lte->staging_list);
-
+               if (lte->unhashed) {
+                       lookup_table_insert(lookup_table, lte);
+                       lte->unhashed = 0;
+               }
+       skip_to_progress:
                do_write_streams_progress(progress,
                                          progress_func,
                                          wim_resource_size(lte));
@@ -791,6 +698,7 @@ do_write_stream_list(struct list_head *my_resources,
 
 static int
 write_stream_list_serial(struct list_head *stream_list,
+                        struct wim_lookup_table *lookup_table,
                         FILE *out_fp,
                         int out_ctype,
                         int write_flags,
@@ -806,7 +714,9 @@ write_stream_list_serial(struct list_head *stream_list,
        progress->write_streams.num_threads = 1;
        if (progress_func)
                progress_func(WIMLIB_PROGRESS_MSG_WRITE_STREAMS, progress);
-       return do_write_stream_list(stream_list, out_fp,
+       return do_write_stream_list(stream_list,
+                                   lookup_table,
+                                   out_fp,
                                    out_ctype, progress_func,
                                    progress, write_resource_flags);
 }
@@ -1274,6 +1184,7 @@ get_default_num_threads()
 
 static int
 write_stream_list_parallel(struct list_head *stream_list,
+                          struct wim_lookup_table *lookup_table,
                           FILE *out_fp,
                           int out_ctype,
                           int write_flags,
@@ -1368,6 +1279,7 @@ out_destroy_res_to_compress_queue:
 out_serial:
        WARNING("Falling back to single-threaded compression");
        return write_stream_list_serial(stream_list,
+                                       lookup_table,
                                        out_fp,
                                        out_ctype,
                                        write_flags,
@@ -1382,18 +1294,19 @@ out_serial:
  * @out_ctype and up to @num_threads compressor threads.
  */
 static int
-write_stream_list(struct list_head *stream_list, FILE *out_fp,
-                 int out_ctype, int write_flags,
-                 unsigned num_threads,
-                 wimlib_progress_func_t progress_func)
+write_stream_list(struct list_head *stream_list,
+                 struct wim_lookup_table *lookup_table,
+                 FILE *out_fp, int out_ctype, int write_flags,
+                 unsigned num_threads, wimlib_progress_func_t progress_func)
 {
        struct wim_lookup_table_entry *lte;
        size_t num_streams = 0;
        u64 total_bytes = 0;
        u64 total_compression_bytes = 0;
        union wimlib_progress_info progress;
+       int ret;
 
-       list_for_each_entry(lte, stream_list, staging_list) {
+       list_for_each_entry(lte, stream_list, write_streams_list) {
                num_streams++;
                total_bytes += wim_resource_size(lte);
                if (out_ctype != WIMLIB_COMPRESSION_TYPE_NONE
@@ -1413,21 +1326,24 @@ write_stream_list(struct list_head *stream_list, FILE *out_fp,
 
 #ifdef ENABLE_MULTITHREADED_COMPRESSION
        if (total_compression_bytes >= 1000000 && num_threads != 1)
-               return write_stream_list_parallel(stream_list,
-                                                 out_fp,
-                                                 out_ctype,
-                                                 write_flags,
-                                                 num_threads,
-                                                 progress_func,
-                                                 &progress);
+               ret = write_stream_list_parallel(stream_list,
+                                                lookup_table,
+                                                out_fp,
+                                                out_ctype,
+                                                write_flags,
+                                                num_threads,
+                                                progress_func,
+                                                &progress);
        else
 #endif
-               return write_stream_list_serial(stream_list,
-                                               out_fp,
-                                               out_ctype,
-                                               write_flags,
-                                               progress_func,
-                                               &progress);
+               ret = write_stream_list_serial(stream_list,
+                                              lookup_table,
+                                              out_fp,
+                                              out_ctype,
+                                              write_flags,
+                                              progress_func,
+                                              &progress);
+       return ret;
 }
 
 struct lte_overwrite_prepare_args {
@@ -1473,7 +1389,7 @@ wim_prepare_streams(WIMStruct *wim, off_t end_offset,
        int ret;
 
        for (int i = 0; i < wim->hdr.image_count; i++) {
-               ret = lte_overwrite_prepare(wim->image_metadata[i].metadata_lte,
+               ret = lte_overwrite_prepare(wim->image_metadata[i]->metadata_lte,
                                            &args);
                if (ret)
                        return ret;
@@ -1482,17 +1398,73 @@ wim_prepare_streams(WIMStruct *wim, off_t end_offset,
                                      lte_overwrite_prepare, &args);
 }
 
+struct stream_size_table {
+       struct hlist_head *array;
+       size_t num_entries;
+       size_t capacity;
+};
+
+static int
+init_stream_size_table(struct stream_size_table *tab, size_t capacity)
+{
+       tab->array = CALLOC(capacity, sizeof(tab->array[0]));
+       if (!tab->array)
+               return WIMLIB_ERR_NOMEM;
+       tab->num_entries = 0;
+       tab->capacity = capacity;
+       return 0;
+}
+
+static void
+destroy_stream_size_table(struct stream_size_table *tab)
+{
+       FREE(tab->array);
+}
+
+static int
+stream_size_table_insert(struct wim_lookup_table_entry *lte, void *_tab)
+{
+       struct stream_size_table *tab = _tab;
+       size_t pos;
+       struct wim_lookup_table_entry *hashed_lte;
+       struct hlist_node *tmp;
+
+       pos = hash_u64(wim_resource_size(lte)) % tab->capacity;
+       lte->unique_size = 1;
+       hlist_for_each_entry(hashed_lte, tmp, &tab->array[pos], hash_list_2) {
+               if (wim_resource_size(hashed_lte) == wim_resource_size(lte)) {
+                       lte->unique_size = 0;
+                       hashed_lte->unique_size = 0;
+                       break;
+               }
+       }
+
+       hlist_add_head(&lte->hash_list_2, &tab->array[pos]);
+       tab->num_entries++;
+       return 0;
+}
+
+
+struct find_streams_ctx {
+       struct list_head stream_list;
+       struct stream_size_table stream_size_tab;
+};
+
 static int
 inode_find_streams_to_write(struct wim_inode *inode,
                            struct wim_lookup_table *table,
-                           struct list_head *stream_list)
+                           struct list_head *stream_list,
+                           struct stream_size_table *tab)
 {
        struct wim_lookup_table_entry *lte;
        for (unsigned i = 0; i <= inode->i_num_ads; i++) {
                lte = inode_stream_lte(inode, i, table);
                if (lte) {
-                       if (lte->out_refcnt == 0)
-                               list_add_tail(&lte->staging_list, stream_list);
+                       if (lte->out_refcnt == 0) {
+                               if (lte->unhashed)
+                                       stream_size_table_insert(lte, tab);
+                               list_add_tail(&lte->write_streams_list, stream_list);
+                       }
                        lte->out_refcnt += inode->i_nlink;
                }
        }
@@ -1502,31 +1474,85 @@ inode_find_streams_to_write(struct wim_inode *inode,
 static int
 image_find_streams_to_write(WIMStruct *w)
 {
+       struct wim_image_metadata *imd;
+       struct find_streams_ctx *ctx;
        struct wim_inode *inode;
-       struct hlist_node *cur;
-       struct hlist_head *inode_list;
+       struct wim_lookup_table_entry *lte;
+
+       ctx = w->private;
+       imd = wim_get_current_image_metadata(w);
+
+       image_for_each_unhashed_stream(lte, imd) {
+               lte->out_refcnt = 0;
+               wimlib_assert(lte->unhashed);
+               wimlib_assert(lte->my_ptr != NULL);
+       }
 
-       inode_list = &wim_get_current_image_metadata(w)->inode_list;
-       hlist_for_each_entry(inode, cur, inode_list, i_hlist) {
+       /* Go through this image's inodes to find any streams that have not been
+        * found yet. */
+       image_for_each_inode(inode, imd) {
                inode_find_streams_to_write(inode, w->lookup_table,
-                                           (struct list_head*)w->private);
+                                           &ctx->stream_list,
+                                           &ctx->stream_size_tab);
        }
        return 0;
 }
 
+/* Given a WIM that from which one or all of the images is being written, build
+ * the list of unique streams ('struct wim_lookup_table_entry's) that must be
+ * written, plus any unhashed streams that need to be written but may be
+ * identical to other hashed or unhashed streams being written.  These unhashed
+ * streams are checksummed while the streams are being written.  To aid this
+ * process, the member @unique_size is set to 1 on streams that have a unique
+ * size and therefore must be written.
+ *
+ * The out_refcnt member of each 'struct wim_lookup_table_entry' is set to
+ * indicate the number of times the stream is referenced in only the streams
+ * that are being written; this may still be adjusted later when unhashed
+ * streams are being resolved.
+ */
 static int
-write_wim_streams(WIMStruct *w, int image, int write_flags,
-                            unsigned num_threads,
-                            wimlib_progress_func_t progress_func)
+prepare_stream_list(WIMStruct *wim, int image, struct list_head *stream_list)
 {
+       int ret;
+       struct find_streams_ctx ctx;
 
-       for_lookup_table_entry(w->lookup_table, lte_zero_out_refcnt, NULL);
-       LIST_HEAD(stream_list);
-       w->private = &stream_list;
-       for_image(w, image, image_find_streams_to_write);
-       return write_stream_list(&stream_list, w->out_fp,
-                                wimlib_get_compression_type(w), write_flags,
-                                num_threads, progress_func);
+       for_lookup_table_entry(wim->lookup_table, lte_zero_out_refcnt, NULL);
+       ret = init_stream_size_table(&ctx.stream_size_tab, 9001);
+       if (ret)
+               return ret;
+       for_lookup_table_entry(wim->lookup_table, stream_size_table_insert,
+                              &ctx.stream_size_tab);
+       INIT_LIST_HEAD(&ctx.stream_list);
+       wim->private = &ctx;
+       for_image(wim, image, image_find_streams_to_write);
+       destroy_stream_size_table(&ctx.stream_size_tab);
+
+       INIT_LIST_HEAD(stream_list);
+       list_splice(&ctx.stream_list, stream_list);
+       return 0;
+}
+
+/* Writes the streams for the specified @image in @wim to @wim->out_fp.
+ */
+static int
+write_wim_streams(WIMStruct *wim, int image, int write_flags,
+                 unsigned num_threads,
+                 wimlib_progress_func_t progress_func)
+{
+       int ret;
+       struct list_head stream_list;
+
+       ret = prepare_stream_list(wim, image, &stream_list);
+       if (ret)
+               return ret;
+       return write_stream_list(&stream_list,
+                                wim->lookup_table,
+                                wim->out_fp,
+                                wimlib_get_compression_type(wim),
+                                write_flags,
+                                num_threads,
+                                progress_func);
 }
 
 /*
@@ -1659,7 +1685,7 @@ finish_write(WIMStruct *w, int image, int write_flags,
        } else {
                memcpy(&hdr.boot_metadata_res_entry,
                       &w->image_metadata[
-                         hdr.boot_idx - 1].metadata_lte->output_resource_entry,
+                         hdr.boot_idx - 1]->metadata_lte->output_resource_entry,
                       sizeof(struct resource_entry));
        }
 
@@ -1671,7 +1697,7 @@ finish_write(WIMStruct *w, int image, int write_flags,
        }
 
        ret = write_header(&hdr, out);
-       if (ret != 0)
+       if (ret)
                goto out;
 
        if (write_flags & WIMLIB_WRITE_FLAG_FSYNC) {
@@ -1789,19 +1815,19 @@ wimlib_write(WIMStruct *w, const tchar *path,
        }
 
        ret = begin_write(w, path, write_flags);
-       if (ret != 0)
+       if (ret)
                goto out;
 
        ret = write_wim_streams(w, image, write_flags, num_threads,
                                progress_func);
-       if (ret != 0)
+       if (ret)
                goto out;
 
        if (progress_func)
                progress_func(WIMLIB_PROGRESS_MSG_WRITE_METADATA_BEGIN, NULL);
 
        ret = for_image(w, image, write_metadata_resource);
-       if (ret != 0)
+       if (ret)
                goto out;
 
        if (progress_func)
@@ -1818,7 +1844,7 @@ static bool
 any_images_modified(WIMStruct *w)
 {
        for (int i = 0; i < w->hdr.image_count; i++)
-               if (w->image_metadata[i].modified)
+               if (w->image_metadata[i]->modified)
                        return true;
        return false;
 }
@@ -1888,7 +1914,6 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags,
        int ret;
        struct list_head stream_list;
        off_t old_wim_end;
-       bool found_modified_image;
 
        DEBUG("Overwriting `%"TS"' in-place", w->filename);
 
@@ -1949,7 +1974,9 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags,
        if (!list_empty(&stream_list)) {
                DEBUG("Writing newly added streams (offset = %"PRIu64")",
                      old_wim_end);
-               ret = write_stream_list(&stream_list, w->out_fp,
+               ret = write_stream_list(&stream_list,
+                                       w->lookup_table,
+                                       w->out_fp,
                                        wimlib_get_compression_type(w),
                                        write_flags, num_threads,
                                        progress_func);
@@ -1959,12 +1986,11 @@ overwrite_wim_inplace(WIMStruct *w, int write_flags,
                DEBUG("No new streams were added");
        }
 
-       found_modified_image = false;
        for (int i = 0; i < w->hdr.image_count; i++) {
-               if (w->image_metadata[i].modified) {
+               if (w->image_metadata[i]->modified) {
                        select_wim_image(w, i + 1);
                        ret = write_metadata_resource(w);
-                       if (ret != 0)
+                       if (ret)
                                goto out_ftruncate;
                }
        }